Vol 17 No 2 - July-December 2017

International Journal of Computer Science
and Business Informatics

(IJCSBI.ORG)
ISSN: 1694-2507 (Print)

VOL 17, NO 2
ISSN: 1694-2108 (Online) JULY-DECEMBER 2017
IJCSBI.ORG
Table of Contents VOL 17, NO 2 JULY-DECEMBER 2017
An Adaptive and Real-Time Fraud Detection Algorithm in Online Transactions .................................... 1

Yiming WU, Siyong CAI
Agronomic Disaster Management using Artificial Intelligence - A Case Study ..................................... 13

M Sudha
Household Power Optimisation and Monitoring System ................................................................... 23

John Batani, Silence Dzambo, Israel Magodi
International Journal of Computer Science and Business Informatics
IJCSBI.ORG
An Adaptive and Real-Time Fraud

Detection Algorithm in Online
Transactions
John Batani
ICT and Electronics Department
Chinhoyi University of Technology, P.Bag 7724, Chinhoyi, Zimbabwe
ABSTRACT
While the Internet has made it possible to transact electronically and ubiquitously, some
unscrupulous internet users have devised ways of defrauding e-commerce users. Several
solutions have been designed and deployed to try and curb fraud in electronic transactions,
but the news of fraud in e-commerce continues making the headlines globally. It is against
this background that the researcher was motivated to design an adaptive algorithm that can
detect credit card fraud as it occurs (real-time). The solution is based on the use of an
Artificial Neural Network, Hidden Markov Model and a One-Time Password. The
researcher used a synthesised dataset since a real dataset could not be found. The researcher
tested the algorithm, which produced 100 per cent fraud detection rate and 98 per cent
accuracy. The proposed solution can be made a plugin to e-commerce sites for the purposes
of detecting and preventing fraud. The researcher was motivated to undertake this study
after realising that while Zimbabwe is calling for the adoption of e-commerce due to the
prevailing cash crisis, some people still have reservations due to security concerns. Despite,
even in those countries where electronic commerce was adopted a long time ago, security is
still a concern among e-commerce participants. The designed algorithm has a learning
ability so that it can detect new fraud variations as they occur (real-time) and thus terminate
the transaction should it be considered a fraudulent one. The author seeks to restore and
instill confidence in people who transact online using credit cards.
Keywords
Fraud detection, Real-time fraud, Adaptive fraud, E-commerce security, Credit card.
1. INTRODUCTION
The Internet has undoubtedly revolutionized the way we do business today.
The increasing popularity of the Internet world over has seen the widespread
acceptance, usage and adoption of electronic commerce in which transacting
can occur without physical interaction, virtually from anywhere across the
globe. However, an upsurge in electronic transactions has presented an
opportunity for unscrupulous computer users to defraud unsuspecting
victims who transact online. Despite several fraud detection solutions
ISSN: 1694-2108 | Vol. 17, No. 2. JULY-DECEMBER 2017 1

IJCSBI.ORG
having been designed and deployed, cases of fraud in electronic transactions
continue making the headlines globally. This suggests that the current
solutions have some weaknesses that are being taken advantage of by
fraudsters. There is sort of an arms race between online fraudsters and those
engaged in designing anti-fraud solutions, hence the need to come up with
an adaptive solution that detects fraud in real time. A good strategy and the
main goal of banks and industries is timely information on fraudulent
activities [3].
.
1.1 The Research Problem
Despite several attempts having been made to curb online fraud, cases of
people being defrauded online continue making the headlines world over.
Notwithstanding the enormous growth in electronic transactions, there is a
lack of strong security to the high end [3]. The rise in Internet usage and an
upsurge in electronic transactions have seen an increase in cases of online
fraud despite significant efforts by card issuers, merchants and law
enforcement to curb the fraud. Since fraud perpetration techniques are
evolving as technology is evolving, there is a need to come up with an
adaptive solution that also has a capability of detecting fraud in real time.
An early detection of fraud is more significant in terms of cost analysis [3];
hence the need to have real time detection solution.
1.2 Purpose Of The Study
The purpose of this study is to understand the shortcomings of current
online fraud detection systems with the aim of designing and implementing
an adaptive, hybrid and real-time online fraud detection algorithm. The
algorithm should have a learning ability so that it can detect new fraud
variations as they occur and thus terminate the transaction should it be
considered a fraudulent one. The algorithm to be designed is supposed to
restore and instill confidence in people who transact online
1.3 Objectives Of The Study
This study seeks to:
1. design an adaptive fraud detection algorithm that can detect credit
card fraud in real time.
2. implement an adaptive fraud detection algorithm that can detect
fraud in real time.
2. RELATED WORKS
“Fraud can be defined as the illegal usage of any system or good” [2], while
the legal activities can correspondingly be termed legitimate. Therefore,
credit card fraud can be defined as the illegal usage of a credit card to
perform electronic transactions. Fraud detection is a complex computational

IJCSBI.ORG
task and there is no system that surely predicts any transaction as fraudulent;
rather the systems predict the likelihood of a transaction being fraudulent
[1]. Basically, there are two types of credit card fraud, that is, counterfeit
fraud and the illegal use of a lost or stolen credit card [2]. Broadly, there are
four types of fraud, namely bankruptcy fraud, theft or counterfeit fraud,
application fraud and behavioural fraud [4]. Several fraud detection
techniques on electronic transactions have been designed, deployed and
applied by various researchers and organisations to reduce further damage
caused by fraud; however, people continue to be defrauded in spite of such
efforts having been made. Despite the existence of numerous fraud detection
technologies, it is not possible to detect fraud while the transaction is in
progress [3], that is, in real time. Some of the fraud detection techniques that
have been implemented include the Hidden Markov Model, decision trees,
neural networks, Bayesian networks and data mining techniques [9].
According to [1], several researches have been done with an emphasis on
data mining and neural networks. [5] applied unsupervised neural networks
in credit card fraud detection. The Hidden Markov Model has previously
been used in credit card fraud detection [10], thus it can be used for credit
card detection. [10] define Hidden Markov Model as a finite set of states,
each of which is associated with a probability distribution. The model only
shows the result and hides the state from the external viewer, thus the states
are „hidden‟ to the outside and hence the name Hidden Markov Model.
Bayesian Networks have also been used in an attempt to detect online fraud
[3]. According to [3], a Naïve Bayesian classifier is an influential
probabilistic method that makes use of class sequence from training class of
prospect instances. Other techniques that have been applied to fraud
detection include self-organizing maps (SOM), K-Nearest Neighbor [3],
Outlier Techniques [6] and also the Boat algorithm. However, regardless of
all such concerted efforts, the news of people being defrauded online
continues making the headlines. This therefore is suggestive that the current
solutions, regardless of them being numerous, have some shortcomings
which are being exploited by online fraudsters. This is therefore what has
prompted the researcher to consider this area as a possible research area.
3. METHODOLOGY
The researcher used the design science research methodology since the
research sought to come up with an artifact. For problem awareness, the
researcher performed document analysis and established that the problem
really exists. The researcher also searched for trends of online fraud in
different presses. Having understood the problem, the researcher proposed
to have a hybrid, adaptive and real-time fraud detection system for credit
card electronic transactions. The researcher then had a tentative design of
the system. The tentative design was then refined, an algorithm designed

IJCSBI.ORG
and system development was undertaken. The system was evaluated for
performance using a test data set, the results of which are shown herein.
This research contributes to the detection of online credit card fraud
detection by proposing a solution that is adaptive and real-time. This is
essential as it ensures that fraud is detected as it happens rather than after the
transaction. Moreover, the use of neural networks ensures that the solution
is adaptive, thus keeping abreast with changes in the ways in which online
credit card fraud is perpetrated. This results in improved security of online
transactions, hence, restoring and boosting confidence of credit card users in
transacting electronically using their cards.
3.1 The Proposed Solution
The researcher designed a hybrid and adaptive fraud detection algorithm
that can detect fraud in real time. It has come out from literature that the
current solutions are not real time. Despite the existence of numerous fraud
detection technologies, it is not possible to detect fraud while the transaction
is in progress [3]. Therefore, the researcher designed an algorithm with the
capability of real time fraud detection. The researcher intends to make use
of neural networks, machine learning and other artificial intelligence
approaches in coming up with the intended solution.
3.2 The Flow Chart of the Proposed System
The user first has to be registered on the e-commerce site, and every time
they try to transact on the site, they are authenticated. After successful
authentication on the website, the user can then enter their credit card details
which will also be verified from the bank database. If the entered credit card
details are incorrect or do not exist, the transaction halts, otherwise it
proceeds. If the credit card details exist in the bank database and are correct,
the system generates a One-Time Password (OTP) which is sent to the
registered mobile number of the cardholder. The user is then prompted to
enter the received OTP and if it matches the sent OTP, then the system
extracts the user‟s social profile from the bank database. This social profile
comprises age, income, occupation, and cardholder‟s value of assets. These
social profile parameters will then be classified and assigned weights using
Artificial Neural Networks to generate the cardholder‟s social status. The
system then extracts credit card and bank transactions history and current
balance (financial profile) from the bank database for analysis using the
Hidden Markov Model (HMM). The HMM is used to generate the
cardholder‟s financial status. Bank transactions involve checking the
validity of the card, card holder‟s previous one-year transactions history,
and balance in the account. Credit card transactions history involves credit
card bill payments and spending patterns of the cardholder. The system uses
the cardholder‟s financial status, social status and OTP for fraud detection.
If authentication is successful (user login credentials, credit card details and
OTP) and the transaction has unique features compared to the cardholder‟s

IJCSBI.ORG
credit card spending history patterns, the user‟s financial profile is updated
using ANNs and the transaction is processed. However, the transaction must
be within the spending limits of the credit cardholder as stipulated by their
bank. If the transaction has new features and exceeds the cardholder‟s
spending limits, it is flagged as fraudulent and terminated.
The continuous updating of the cardholder‟s financial profile through ANNs

makes the fraud detection system adaptive, while the use of OTP makes it a
real-time fraud detection system. Combining ANNs and the HMM makes
the system a hybrid solution.
Figure 1. System flow chart

IJCSBI.ORG
3.3 Algorithm for the Proposed System

Let InT be incoming transaction, PrT be previous transactions and OTP be
one-time password, FrT be fraudulent transaction and LgT be legitimate
transaction
Steps
I. Receive login details
II. If login details are valid Then
a. Send OTP
b. Compare received OTP and sent OTP
i. If received OTP !=sent OTP Then
1. Terminate transaction and End
ii. Else
1. Accept credit card details
2. Authenticate credit card details
3. If credit card details are invalid Then
a. Terminate Transaction & flag it as
suspicious
4. Else
a. Extract cardholder‟s social profile
{name, age, gender, income,
occupation, value of assets}
b. If cardholder name or gender or age
!=user‟s name or gender or age Then
i. Terminate the transaction and
flag it as fraudulent
ii. End
c. Extract cardholder‟s financial profile
{card balance, card transactions
history, card bills history, card
payments history, credit limit}
d. If InT value > credit limit Then
i. Flag InT as fraudulent
ii. End
e. Classify social profile parameters
using ANNs and assign them weights
f. Analyse financial profile using HMM
to generate financial status
g. If InT has new features
i. Update cardholder‟s spending
patterns using ANNs
h. If InT resembles PrT Then
i. InT is LgT

IJCSBI.ORG
ii. Commit InT to legitimate
transactions database
iii. LgT + =1
iv. Notify user and bank of the
transaction
v. End
i. Else
i. InT is fraudulent
ii. Record transaction as FrT
iii. FrT += 1
iv. Notify cardholder and bank
v. End
III. Else
a. Terminate transaction
IV. END
Neural Network Algorithm

Let:
InT be incoming transaction
LegT be legal transaction
FrT be fraudulent transaction
SusT be suspicious transaction
MaxOTP_Time be the maximum number of seconds within which a
user is supposed to enter an OTP
Time_Elapsed be time in seconds that has elapsed after an OTP has
been sent and before user has entered the corresponding OTP.
InT_Features{}be a set of features for incoming transaction for
customer
LegT_Features{} be a set of features for legal transaction for
customer
FrT_Features {} be a set of features for fraudulent transactions for
customer
NewFeatures{} be a set of new features from incoming transaction
that do not exist in the current dataset of legal transactions and
fraudulent transactions
Define Variables:
Threshold {user defined variable in the form of a percentage, which
is used to compare similarity between incoming Transaction and

IJCSBI.ORG
legal transaction for each customer. It can be between 0 and 1
inclusive, or a percentage}
Step 1: Extract attributes of InT and store them in InT_Features{}
Step 2: Compare InT_Features{} and LegT_Features{}
If similarity >= threshold Then
InT is LegT
Else
InT is SusT
Suspend InT
Send OTP to cardholder‟s mobile number
If Time_Elapsed = MaxOTP_Time Then
InT is FrT
Extract new features from InT and update
FrT_Features{} for customer
Else
InT is LegT
Extract new features from InT and update
LegT_Features{} for customer
End If
End If
4. RESULTS
The results of the system were evaluated in terms of the extent to which it
correctly classified fraudulent transactions. In other words, the system was
mainly evaluated on fraud detection rate. The system was tested using
synthetically generated data since the researcher could not get a real dataset.
However, the data was synthesised in a way that it as much as possible
resembled a real dataset. Herein, a positive is a fraudulent transaction and
conversely, a negative is a legal transaction. Thus, a legal transaction that is
wrongly classified as fraudulent is a false positive; and a fraudulent
transaction that is misclassified as legal is a false negative. Therefore, legal
transactions that are correctly classified as legal are true negatives, and
fraudulent transactions that are correctly classified as fraudulent are true
positives. The system was thus evaluated in line with these parameters.

IJCSBI.ORG
Table 1. Training Results
No. of No. No. of No. of No. of No. of No. of No. of
Iterations of False True False True Positiv Negati
Trans Positiv Positi Negati Negati es ves
action es ves ves ves
s
50 50 2 25 0 23 25 25
100 50 0 25 0 25 25 25
150 50 1 25 0 24 25 25
Table 2. Testing Results

No. of No. of No. of No. of No. of No. of No. of
Transac False True False True Positive Negati
tions Positives Positive Negative Negatives s ves
s s
50 1 25 0 24 25 25
From the testing results, the fraud detection rate can thus be calculated using
the formula:

IJCSBI.ORG
Sensitivity is regarded as the most important measure of the effectiveness of

fraud detection systems [8]. This is because losses of money by defrauded
people are a result of fraud detection systems failing to detect fraud. While
this system‟s detection rate was perfect, it should be noted that its
perfectness depends on some factors. Firstly, the system is deployed to work
with electronic commerce transactions on which users are required to
register. The system blocks all transactions which are initiated by users
whose demographic details are different from the cardholder‟s as stored in
the bank‟s database. For a transaction to be successful, the registered details
on the website for the user must match with those for the cardholder, and the
initiator of the transaction must have the cardholder‟s mobile phone to
which an OTP will be sent. If all those conditions are met, a transaction will
sail through even if some of its features are not found in the legal
transactions set for that customer. The new features of an incoming
transaction will then be added to either the legal transactions or fraudulent
transactions features set for the customer, thus making the solution adaptive
or incremental since it has an ability to learn new features associated with
fraud.
The system does not only detect but also prevents fraud by prematurely
terminating any transaction should the user fail to enter the generated OTP
which is sent to their registered mobile phone number via Short Messaging
System (SMS). Consequently, it achieves both credit card detection and
prevention.
In comparison to other systems in existence, the proposed algorithm is

highly effective with 100 per cent fraud detection rate. According to [7], the
most effective credit card fraud detection system, which is the Dempster and
Shafer Theory and Bayesian Learning, has 98 per cent fraud detection rate.
Bayesian and Neural Network, Hidden Markov Model, and Hybridization
of BLAST-SSAHA have fraud detection rates of 77, 70 and 86 percent
respectively [7]. Consequently, the algorithm proposed herein stands out as
the best. In terms of accuracy, the proposed algorithm has a very high
accuracy of 98 per cent. However, [7] do not specify the numerical accuracy
of Dempster and Shafer Theory and Bayesian Learning; Bayesian and
Neural Network; Hidden Markov Model; and Hybridization of BLAST-
SSAHA but simply specify their accuracy as high, medium, medium and
high, respectively.

IJCSBI.ORG
5. CONCLUSIONS
The proposed solution to fraud detection puts more emphasis on ensuring
customer security in electronic transactions using credit cards. The strengths
of the proposed algorithm include a high accuracy of 98%, a high fraud
detection rate of 100%, ability to learn new fraud variations and new
customer spending patterns (adaptive), and ability to detect and prevent
fraud as it occurs (real-time). While the system has a high fool proof or
fraud detection rate, it should be noted that a customer cannot transact even
if they are the cardholder unless they have their registered mobile phone
number and have registered their details on the website just as they are
captured in the credit card issuer‟s database. The researcher‟s view is that
security matters more than convenience, hence one cannot transact without
their registered mobile number. An OTP is sent to a mobile phone via text
messaging and not by electronic mail (e-mail) since e-mail can be easily
hacked. The other reason for choosing to send an OTP via SMS was that a
recipient is not charged for receiving an SMS in the author‟s country of
residence. Prospective users of this algorithm may also decide to send an
OTP to an e-mail rather than via SMS. The effectiveness of the proposed
algorithm may be compromised if a criminal gets access to the legitimate
cardholder‟s registered phone number, and e-commerce website
authentication credentials.
REFERENCES
[1] Dheepa V. and Dhanapal R., "Analysis of Credit Card Fraud Detection Methods,"
International Journal of Recent Trends in Engineering, vol. 2, no. 3, pp. 126-128,
2009.
[2] Ekrem D. and Hamdi O.M, "Detecting credit card fraud by genetic algorithm and
scatter search," EXPERT SYSTEMS WITH APPLICATIONS, vol. 38, no. 10, pp.
13057-13063, 2011.
[3] Gayathri R. and Malathi A., "Investigation of Data Mining Techniques in Fraud
Detection: Credit Card," International Journal of Computer Applications, vol. 82, no.
9, pp. 12-15, 2013.
[4] Linda D., Hussein A., and Pointon J., "Credit card fraud and detection techniques: a
review," Banks and Bank Systems, vol. IV, no. 2, pp. 57-68, 2009.
[5] Ogwueleka F.N, "Data mining applications in credit card fraud credit card fraud
detection system," Journal of Engineering Science and Technology, vol. 6, no. 3, pp.
311-322, 2011.
[6] Rama K. K and Uma D. D., "Fraud Detection of Credit Card Payment System by
Genetic Algorithm," International Journal of Scientific & Engineering Research, vol.
3, no. 7, pp. 1-6, 2012.
[7] Rana P.J and Baria J., "A Survey on Fraud Detection Techniques in Ecommerce,"
International Journal of Computer Applications, pp. 5-7, 2015.
[8] Seeja K.R and Zareapoor M., "FraudMiner: A Novel Credit Card Fraud Detection
Model Based on Frequent Itemset Mining," Scientific World Journal, vol. 2014, 2014.

IJCSBI.ORG
[9] Sharma A. and Panigrahi P.K, "A review of financial accounting fraud detection based
on data mining techniques," International Journal of Computer Applications, vol. 39,
no. 1, pp. 37-47, 2012.
[10] Singh A. and Narayan D., "A Survey on Hidden Markov Model for Credit Card Fraud
Detection," International Journal of Engineering and Advanced Technology, vol. 1,
no. 3, pp. 49-52, 2012.
This paper may be cited as:

Batani, J. 2017. An Adaptive and Real-Time Fraud Detection Algorithm in
Online Transactions. International Journal of Computer Science and
Business Informatics, Vol. 17, No. 2, pp. 1-12.

IJCSBI.ORG
Agronomic Disaster Management

using Artificial Intelligence -
A Case Study
M. Sudha
Associate Professor
School of Information Technology and Engineering
VIT - India
ABSTRACT
Artificial Intelligence has become an essential tool in various hydrological data-driven
forecast scenarios. The existing needs of farm management activities have witnessed the
necessity of an intelligent decision support for strategic planning and implementation. This
case study reports the benefits of application of neural network based short range
precipitation prediction model in agronomic disaster management (ADM). This
investigation describes the benefits of data-driven decision support systems for agronomic
sustainability. Neural Network Architecture emerged recently have heterogeneous network
design thereby suits for solving complex problems. The methodical evaluation conducted
on agronomic disaster management framework designed using rough, genetic and neuro
computing approach reported peak prediction accuracy of 97.21 % while the learning rate
of the network was set to 0.7 for a fixed momentum of 0.5 producing a nominal error rate
of 02.79%.
Keywords
Artificial intelligence, Agronomic disaster management, Daily precipitation prediction,
Data-driven computing.
1. INTRODUCTION
As a recent trend rapid proliferation of computers and developments in the
area of information technology has increased the applications of
computational intelligence in meteorological aspects. Computational
intelligence approach has exposed a significant amendment in the
advancement of new hybrid data-driven techniques in modeling the wide
range of weather data [1], [2], [3] and [4]. Data-driven models are simple to
understand and require no prior experience or expertise. Decision making is
an execution of an activity that is more or less a human quest. A decision
support system is a computer-based information system that help decision
maker to deal with the problems through direct interaction with data and
analysis model [5]. Soft computing approaches incorporate efficient
computational methodologies stimulated by intrinsic vagueness, intuition
and acquaintance of rational thinking and real world uncertainty. The
intelligent systems designed to handle uncertainty in real life problems

IJCSBI.ORG
usually make use of rough set theory, neural network, evolutionary

algorithms, and fuzzy sets approaches [6] and [7] Fuzziness and uncertainty
exist almost everywhere, it is inevitable in atmospheric dynamics.
2. RELATED WORKS
A systematic review was conducted to assess the applicability of data
mining, artificial neural network, fuzzy inference system and evolutionary
approaches in the modeling hydrological forecast. The review of the
presented literature revealed that more often than rare the models have
shown average performance in modeling hydrological predictions when
compared with techniques that employ the concept of assimilating one or
more intelligent approaches. From the following detailed review, it is clear
that there is a necessity of much better hybrid soft computing approach
which can handle weather parameters more intelligently rather than a black
box model. In [8] it has been reported ANN as a superior method for
modeling precipitation forecast scenarios. In [9], ANN has been applied for
modeling daily precipitation forecasting in the Mashhad synoptic station.
In [10] they have applied ANNs to predict the month wise maximum
precipitation, minimum precipitation, average and total cumulative
precipitation during a period of the next four consecutive months. [11]
compared the performance of General Regression Neural Network (GRNN),
ensemble neural network, BPNN, Radial Basis Function Network (RBFN),
GA, MLP and fuzzy clustering for precipitation prediction. [12] applied
ANN models and decision tree algorithm to forecast maximum temperature,
precipitation, evaporation and wind speed at Ibadan city located in Nigeria.
The results concluded ANN as a suitable tool for meteorological
predictions. ANN and Fuzzy Logic (FL) algorithm for forecasting the
stream-flow for the catchment of Savitri River Basin using 20 years (1992–
2011) precipitation and other hydrological data. While comparing both
ANN and FL algorithms the investigation and empirical results reported that
for prediction of hydrological scenarios ANN performance is quite superior
to FL[8].
Assimilating the features of ANN and FIS has attracted the rising attention
of researchers due to the growing requisite of adaptive intelligent systems to
solve the real world requirements [13]. [14] ANFIS models are widely used
in modeling daily precipitation prediction. [15] Developed a Modified
ANFIS for modeling the nonlinear dynamic characteristics of precipitation
events at the Klang River basin; Malaysia. [16] developed two different
models for forecasting weather at two separate regions, Amman airport and
Taipei, China using artificial neural networks and fuzzy logic. The
prediction accuracy achieved by proposed models was satisfactory. [17]

IJCSBI.ORG
stated that ANFIS based models developed were configured and evaluated
for six major dams of South Korea having high, medium and low reservoir
capacity. The results showed significant improvement for categorical
precipitation forecast using ANFIS. [18] applied ANFIS for forecasting
drought, the quantitative value of drought indices and the Standardized
Precipitation Index (SPI).
In [6], a neuro-fuzzy model was developed to predict the monthly

precipitation of the Daejeon Station in Korea. Choubin et al. (2014)
developed a neuro-fuzzy model to forecast annual drought conditions in the
Maharlu-Bakhtegan watershed, located in Iran. The study reported neuro-
fuzzy model as a suitable method to analyze the influence of independent
variables on dependent variables in hydrological applications. An ANFIS
model has been applied to forecast of the groundwater level of Bastam Plain
in Iran [19]. [20] Developed ANFIS for modeling long-term streamflow
forecasting in Dez basin, Iran. The results reported ANFIS as a suitable
method for streamflow forecasting, and K-fold cross validation method
could increase the model reliability. [14] Reinstated that the performance of
fuzzy inference system and artificial neural network based soft computing
techniques are better than the L-moments approach used for flood
forecasting. [21] Stated that ANFIS perform groundwater level prediction
more accurately when compared to artificial neural networks and Bayesian
neural networks.
3. CASE STUDY REGION

Western-Ghats region of Tamil Nadu State (Coimbatore zone) in India is the
case study region for the assessment of precipitation prediction. This region
serves as Manchester of South India lying in the extreme western part of
Tamil Nadu This district’s total area covers 746,800 hectares, and 43% of
the region is bound to agricultural cultivation. The cotton, sugarcane, peanut
sorghum, maize, rice, and pulses are the primary crops in this area. The
study region is one of the most important agricultural and industrial areas in
the country. Fast and independent industrial development projects have
caused climatological changes in past years, hence raised necessity to
conduct the assessment of factors influencing the current weather prediction.
4. MATERIALS AND METHODS

The day wise observatory record of eight atmospheric parameters for
precipitation prediction, measured in millimeter (mm), for 29 years from
1984 to 2013. The target sample data set used as input for this study is
represented in Table 1, the dataset having 10,000 records with no missing
values and outliers after pre-processing phase is subject to experimental
evaluation. The decision attribute (Pp) is a binary decision variable, Pp =

IJCSBI.ORG
‘no’ denotes no precipitation (rainfall) otherwise if Pp = ‘yes’ indicates

precipitation (rainfall) occurrences. The parameter's values are recorded in
an observatory in the daily basis as per the standard norms.
Table 1. Observatory weather Dataset

Observatory weather Units Data type Data
parameters range
Maximum temperature (Max) Celsius Float 27 - 398
Minimum temperature (Min) Celsius Float 2 - 335
Relative humidity 1 (Rh1) Percentage Float 5 - 100
Relative humidity 2 (Rh2) Percentage Float 1- 99
Wind speed (Ws) Km/hrs Float 01 - 227
Solar radiation (Sr) KCalories Float 24-688
Sunshine (Ss) Hrs Float 01-98
Evapotranspiration (Evp) mm Float 01-75
Precipitation prediction (Pp) mm Categorical Yes/no
5. AGRONOMIC DISASTER MANAGEMENT MODEL

The sequentially hybridized artificial intelligence centered precipitation
prediction model consists of input assessment, model training and testing
phase. The model output enable the framing decisions such harvesting,
sowing seeds, spraying pesticides or fertilizers to crops in the cultivated
regions and other decisions such as storage, sale etc. Henceforth this model
is referred as an agronomic disaster management model. A farmer is assisted
to make a strategic decision on the farming activities such as harvesting
early before damage if there is heavy precipitation in the next one or two
days. Similarly farmers can plan for spraying the fertilizers ahead or in later
time based on the environmental factor owing to precipitation.
This decision support system modelled for agronomic disaster management

is mainly designed based on data-driven modeling concept using rough sets
based evolutionary computing and the three layered multi-layered back-
propagation system. The learner is subject to learn the environment during
the training process from the target input and evaluated with set of test data.
The training algorithm uses sigmoid transfer function which is a well-
known suitable neural network activation function. The input data can be
one of the factors that may influence the output of the architecture.
Subsequently, a multi-layered back-propagation multi-layered algorithm
[22] is employed to train the ADM model in more effective way.

IJCSBI.ORG
The pseudocode of the multi-layered back-propagation algorithm is as

below:
1. Begin
2. Initialize with randomly chosen weights and biases in network;
3. while error is above the threshold, do
4. for each training tuple X in W
5. {
6. for each input layer unit k
7. {
8. for each hidden or output layer unit k
9. {
10.
11. //compute the net input of unit k through previous layer, i
12. for each unit k in the output layer
13. 𝐸𝑟𝑟𝑗 = 𝑂𝑘(1 − 𝑂𝑘) (𝑇𝑘 − 𝑂𝑘)
14. 𝐸𝑟𝑟𝑗 = 𝑂𝑘(1 − 𝑂𝑘) 𝛴𝑗 𝐸𝑟𝑟𝑗 𝑊𝑘𝑗
15. for each weight in network "n"
16. △ 𝑊𝑖𝑘 = (𝑙)𝐸𝑟𝑟𝑘𝑂𝑖)
17. // weight increment
18.
19. // weight update
20. for each bias in network "n"
21.
22. // bias increment
23.
24. } // bias update
25. } }
The proposed system is developed and implemented using Microsoft

NET framework and its process flow is shown in Fig. 2. It consists of
feature reduction stage, training and testing phase. The target data in the
input for selection stage then the proposed system has been trained by
reducts dataset generated using the proposed technique. The complexity of
training the network for given |D| tuples and w weights, each epoch requires
O (|D| x w) time [22].

IJCSBI.ORG
Task relevant input

data
(1) BP- Perceptron model

(2) Evolutionary Computing
(3) Fuzzy Rule based model
Obtained error ≤ Desired

error
If Accuracy No
attained ≥ Peak
accuracy
Yes
Attained Prediction
Accuracy
Figure 1. Flow design of Agronomic Disaster Management model
5. RESULTS
The learning rate and momentum are set for approximately suitable random
value and adjusted according to attain the desired output. From this
methodical evaluation, the learning rate and momentum is fixed as 0.7 and
0.5 as in Table 2. The comparative evaluation of the various models under
this investigation as projected in Table 3 and Figure 2 evidently reports that
the prediction techniques have substantial improved when trained after
feature reduction and the proposed model acquired high accuracy. The
classification models used for training have shown better results when
trained using the selected list of observatory parameters. The proposed
model outperformed the existing approaches by reporting a nominal error
rate of 2.71 % when compared to other existing classification algorithms.
Random-forest classifier has reported low accuracy of 81.07 % with an
error rate of 19.93 % and revealed the limitation of adopting this classifier
in modeling real-time forecasting. Also, some hidden inferences such has

IJCSBI.ORG
any classification model utilizing artificial intelligence exhibits better

optimization potential when compared to other learning methods. The
learning potential of a neural network and the benefit of error propagation
strategy of the multi-layered back-propagation learning algorithm have
enabled to attain the desired accuracy. The predictive models were judged
statistically using the percent error of prediction and prediction yields. The
experimental assessment revealed the potential of AI centered precipitation
prediction model for harvesting, sowing seeds, spraying pesticides or
fertilizers to crops over other existing prediction models.
Table 2. Performance evaluation of proposed system for adaptable learning rate

Learning Momentum Accuracy Error
Rate rate rate
0.1 0.5 93.67% 6.32 %
0.5 0.5 94.17% 5.82%
0.7 0.5 97.21% 2.79%
Table 3. Performance evaluation of Data-driven AMD vs existing models

Parameter Selection Prediction Models Software Accuracy
Exhaustive-forward Random-forest Weka-tool 81.07 %

selection search classifier
Information gain Fuzzy-unordered- Weka-tool 84.97 %
rule-induction
PSO Bayesian-network Weka-tool 89.75 %

classifier
Attribute weighting Fuzzy-neural Weka-tool 91.79 %
classifier
Relative fitness function Proposed -ADM RSES2.0 97.21
.NET

IJCSBI.ORG
Attained Prediction Accuracy (%)

100
97.21
95
91.79
90 89.75
85 84.97
80 81.07
75
Accuracy (%)
70
Figure 2. Precipitation prediction accuracy optimization report
6. CONCLUSIONS
Modeling user interactive agronomic disaster management model to support
farming activities such as harvesting, sowing seeds, spraying patricides or
fertilizer’s to crops utilizing real-time weather data is the primary concern of
this research. Many intelligent techniques are employed in modeling
precipitation forecast scenario. However, most of the techniques are not
dealing with qualitative data. Therefore, this research has evolved input
selection methods that could remove the superfluous parameter. Besides
this, an efficient data-driven intelligent system is designed assimilating
rough set based evolutionary computing and neural network to attain
optimal prediction accuracy. Also, the limitations of the existing data
mining models and the benefits of intelligent techniques in modeling
weather forecast scenario has been reported based on the experimental
outcomes. The proposed ADM achieved an optimal prediction accuracy of
97.21 % with the nominal error rate of 02.79%. The research attained the
intent of developing an ADM with interactive user interface to enable
operators to make regional prediction to support farming decisions.

IJCSBI.ORG
REFERENCES
[1] Malik, A. and Kumar, A., 2015. Pan evaporation simulation based on daily
meteorological data using soft computing techniques and multiple linear regression.
Water resources management, 29(6), pp.1859-1872.
[2] Sudha, M. and Subbu, K., 2017. Statistical Feature Ranking and Fuzzy Supervised
Learning Approach in Modeling Regional Rainfall Prediction Systems. AGRIS On-line
Papers in Economics and Informatics, 9(2), p.117.
[3] Sudha, M., 2017. Intelligent decision support system based on rough set and fuzzy
logic approach for efficacious precipitation forecast. Decision Science Letters, 6(1),
pp.95-106.
[4] Mohankumar, S. and Balasubramanian, V., 2016. Identifying Effective Features and
Classifiers for Short Term Rainfall Forecast Using Rough Sets Maximum Frequency
Weighted Feature Reduction Technique. CIT. Journal of Computing and Information
Technology, 24(2), pp.181-194.
[5] Sprague Jr, R.H. and Watson, H.J., 1996. Decision support for management. Prentice-
Hall, Inc..
[6] Jeong, C., Shin, J.Y., Kim, T. and Heo, J.H., 2012. Monthly precipitation forecasting
with a neuro-fuzzy model. Water resources management, 26(15), pp.4467-4483.
[7] Pant, L.M. and Ganju, A., 2004. Fuzzy rule-based system for prediction of direct action
avalanches. Current science, 87(1), pp.99-104.
[8] Kothari, M. and Gharde, K.D., 2015. Application of ANN and fuzzy logic algorithms
for streamflow modelling of Savitri catchment. Journal of Earth System Science,
124(5), pp.933-943.
[9] Khalili, N., Khodashenas, S.R., Davary, K., Baygi, M.M. and Karimaldini, F., 2016.
Prediction of rainfall using artificial neural networks for synoptic station of Mashhad: a
case study. Arabian Journal of Geosciences, 9(13), p.624.
[10] Moustris, K.P., Larissi, I.K., Nastos, P.T. and Paliatsos, A.G., 2011. Precipitation
forecast using artificial neural networks in specific regions of Greece. Water resources
management, 25(8), pp.1979-1993.
[11] Narvekar, M. and Fargose, P., 2015. Daily weather forecasting using artificial neural
network. International Journal of Computer Applications, 121(22).
[12] Olaiya, F. and Adeyemo, A.B., 2012. Application of data mining techniques in weather
prediction and climate change studies. International Journal of Information
Engineering and Electronic Business, 4(1), p.51.
[13] Abraham, A., 2001. Neuro fuzzy systems: State-of-the-art modeling techniques.
Connectionist models of neurons, learning processes, and artificial intelligence,
pp.269-276.
[14] Kumar, R., Goel, N.K., Chatterjee, C. and Nayak, P.C., 2015. Regional flood
frequency analysis using soft computing techniques. Water resources management,
29(6), pp.1965-1978.
[15] Akrami, S.A., El-Shafie, A. and Jaafar, O., 2013. Improving rainfall forecasting
efficiency using modified adaptive Neuro-Fuzzy Inference System (MANFIS). Water
resources management, 27(9), pp.3507-3523.
[16] Al-Matarneh, L., Sheta, A., Bani-Ahmad, S., Alshaer, J. and Al-oqily, I., 2014.
Development of temperature-based weather forecasting models using neural networks
and fuzzy logic. International journal of multimedia and ubiquitous engineering,
9(12), pp.343-366.
[17] Awan, J.A. and Bae, D.H., 2014. Improving ANFIS based model for long-term dam
inflow prediction by incorporating monthly rainfall forecasts. Water resources
management, 28(5), pp.1185-1199.

IJCSBI.ORG
[18] Bacanli, U.G., Firat, M. and Dikbas, F., 2009. Adaptive neuro-fuzzy inference system
for drought forecasting. Stochastic Environmental Research and Risk Assessment,
23(8), pp.1143-1154.
[19] Emamgholizadeh, S., Moslemi, K. and Karami, G., 2014. Prediction the groundwater
level of Bastam plain (Iran) by artificial neural network (ANN) and adaptive neuro-
fuzzy inference system (ANFIS). Water resources management, 28(15), pp.5433-5446.
[20] Esmaeelzadeh, S.R., Adib, A. and Alahdin, S., 2015. Long-term streamflow forecasts
by Adaptive Neuro-Fuzzy Inference System using satellite images and K-fold cross-
validation (Case study: Dez, Iran). KSCE Journal of Civil Engineering, 19(7), p.2298.
[21] Maiti, S. and Tiwari, R.K., 2014. A comparative study of artificial neural networks,
Bayesian neural networks and adaptive neuro-fuzzy inference system in groundwater
level prediction. Environmental earth sciences, 71(7), pp.3147-3160.
[22] Han, J., Pei, J. and Kamber, M., 2011. Data mining: concepts and techniques. Elsevier.

Sudha, M. 2017. Agronomic Disaster Management using Artificial
Intelligence - A Case Study. International Journal of Computer Science and

IJCSBI.ORG
Household Power Optimisation and

Monitoring System
John Batani, Silence Dzambo and Israel Magodi
ICT & Electronics Department
Chinhoyi University of Technology,
P.Bag 7724 Chinhoyi, Zimbabwe
ABSTRACT
Like most of the developing countries, Zimbabwe continues to face critical electricity
shortages. In this paper, the authors present a smart solution for reducing electricity usage
in households while improving comfort levels for the dwellers. The authors developed a
prototype to optimise electricity usage by domestic appliances. In attaining optimal power
usage in households and remote manipulation of household appliances, the researchers
utilized the design science research methodology. The proposed system reduced power
usage and cost of electricity in households by at least 50 per cent. The results benefit the
nation by reducing domestic electricity usage and thus reducing the overall electricity
shortages which may affect the manufacturing and other sectors of the economy. Remote
manipulation of, and communication with the devices by the user are achieved through the
Bluetooth technology and the Global System for Mobile communication. When the user
turns on a stove, the system automatically turns off the fridge to minimise power consumed.
However, the performance of this system may be affected by the performance of the
sensors used. The focus of this paper was to reduce the amount of electricity consumed by
households, thus reducing the overall stress on the national power grid and increasing the
available electricity for industrial use, leading to employment creation. The proposed
system can help in the realization of the UN’s SDGs through ensuring there is adequate
electricity for industrial use, as espoused by SDG 9.
Keywords
Power optimization, Household power saving, Electricity saving, Electricity shortage,
SDGs, IT4D, ICT4D.
1. INTRODUCTION
The importance of electricity in any economy cannot be underestimated
[1,2], hence the need to efficiently use it. Zimbabwe currently has a shortage
of electricity and since 2007 the nation has experienced load shedding due
to inadequate generation of electricity by the national power utility company
[3]. [4]has it that there is generally shortage of electricity globally, and
Zimbabwe is no exception. According to [3] Zimbabwe will continue to
have electricity shortages for upto 8 more years due to the incapacity to
generate sufficient electricity. Several attempts, such as the use of energy
savers, electricity importation and use of alternative, natural power sources

IJCSBI.ORG
such as gas and solar, have been made to ease pressure on the insufficient
electricity in Zimbabwe. However, the problem of electricity shortage in
Zimbabwe still persists, and there is a call to everyone to contribute towards
solving the problem [3]. In light of this, the researchers developed a
household power optimization and monitoring system for optimizing the
usage of the inadequate electricity that is currently generated in the country
while not inconveniencing the users.
Africa’s electricity shortage is hugely characterised by continuing power
cuts and a complete deficiency of electricity infrastructure [5]. This has
resulted in negative effects to human and socioeconomic development
across the continent [5]. According to [5], only an average of 40 per cent of
Africans enjoy a consistent electricity supply; while only 69 per cent of the
electrified homes really have electricity that works most or all of the time.
62 per cent of Zimbabwe’s population has access to an electricity grid [5].
[5] claim that only 30 per cent of Zimbabweans have electricity that works
reliably, 26 per cent have electricity that works half the time while 44 per
cent have electricity that either works ocassionally or not at all. Zimbabwe
Electricty Supply Authority (ZESA) is the sole producer, distributor and
seller of electricity. [6] states that the electricity industry in Zimbabwe has
operated as a controlled monolopoly for about five decades. [3] has it that
they will reduce electricity generation from 750MW to 475MW due to
reduced dam levels since most of the electricity in the country in hydro
generated. Zimbabwe has had an 80 per cent urban electrification, 20 per
cent rural electrification, and 41 per cent overall elecrification growth from
1980 to 2007 [7]. An unmatched increasing population and balooning
number of electric appliances has created an electricity shortage in
Zimbabwe, resulting in substantial load shedding [3]. This electricity
shortage is despite several efforts that have been made to increase electricity
supply and reduce electricity consumption in Zimbabwe, including power
importation, use of energy savers and use of alternative energy sources such
as solar and biogas [3]. All over the world, several IT based systems have
been developed in an attempt to reduce excessive power demand, such as
the Green Building in Italy [8]. In Zimbabwe, little has been done to
optimise power usage through the use of ICTs [1]. In 2012 ZESA
introduced pre-paid meters as to enable customers to manage their
electricity bills and encourage them to use electricity wisely. Despite all
these efforts, Zimbabwe still faces electricity shortage [3], hence the need to
come up with a solution for optimisation

IJCSBI.ORG
1.1 Problem Statement
There is generally a serious shortage of electricity in the whole world in
general and Zimbabwe in particular [4]. Zimbabwe has a reliable electricity
capacity of the order of 1 320 MW (Megawatts) against a demand of about 2
200 MW [9]. [10] concurs that Zimbabwe is facing critical electricity
shortages due to inadequate electricity generation. Several efforts have been
made in Zimbabwe to reduce electricity consumption and improve
electricity supply. Such efforts include the use of energy savers, electricity
importation and use of alternative, natural power sources such as gas and
solar. However, Zimbabwe continues to face electricity shortages as
evidenced by massive power cuts and load shedding [3]. This has affected
both domestic and industrial consumers; hence the need to come up with a
solution that optimises electricity usage in Zimbabwe.
1.2 Research Objectives
1. To design an automated system that optimises electricity usage in
households.
2. To design an android application that enables remote manipulation
and monitoring of plugged on household electric appliances.
1.3 Significance of the Study
This study seeks to come up with a solution for optimising electricity usage
in households; hence reducing the load on the national grid. Reducing
electricity consumed by households reduces the national demand for
electricity and may save the country foreign currency in reducing electricity
imports. Moreover, if domestic electricity consumption is reduced, it
increases the amount of electricity available for industrial use, which in turn
may improve employment creation. According to [11], every single
occupation in the manufacturing sector generates more than two million
occupations in other sectors, hence it is important to ensure that the
manufacturing sector is sufficiently powered. In addition to power usage
optimisation, the study also seeks to reduce electricity bills for household
consumers
2. RELATED WORKS
Power optimization refers to reducing the amount of power consumed by
devices (such as home appliances, while preserving their functionality)
through designing automation tools that minimise power wastage [12]. A
well designed monitoring system should be capable of maintaining preset
environmental conditions in the building [13]. Various solutions for energy
saving in households using smart technology have been proposed and
developed. Most approaches in literature for energy saving in households
focus on lowering the power consumed by heating, ventilation and air
conditioning (HVAC) appliances, such as the household heating systems

IJCSBI.ORG
[14], air-conditioning [15] or both of them [16]. [17] further identified
lighting and home appliances as two more areas to be incorporated in energy
management features to minimize the domestic energy waste. Other power
optimisation solutions indirectly attend to lowering the power consumed by
such (HVAC) devices by providing improved monitoring and controlling
options for the devices, which will in turn result in power consumption
being lowered [18]. The majority of such solutions employ a wide range of
sensors for measuring humidity and temperature, the data from which are
processed by fuzzy controllers [14].
[8] designed an automated power management system called the

GreenBuilding. This system used sensors to intelligently monitor power
usage and automatically control the behaviour of devices in a building. The
system provides a dashboard through which a user can view power
consumption statistics by each appliance [8]. The Arduino platform was
used in designing the network of sensors. Although GreenBuilding allows
the user to view reports based of power consumptions, it does not send
notifications/alerts to the user through the phone pertaining electric
appliance consumption or status notifications. Moreover, once
GreenBuilding is installed, it does not allow users to create their own modes
based on their own priority preferences on appliances.
The Smart Grid is another power optimisation solution which is an
amalgamation of communication and electric infrastructure through IT in
the current electrical networks to boost efficiency [19]. This system can
control daily used household devices according to user defined tariff rates
for each particular device, thus reducing electricity costs to the consumer
and reducing pressure on the grid [20].
[21] designed a simple system for remotely controlling and monitoring
lights, using the Global System for Mobile Communication for long range
communication and Bluetooth technology for short range communication.
The system sought to reduce electricity consumed by household devices
through the use of infrared sensor. Apart from reducing electricity usage, the
system also notified users of any irregular situations (like high temperatures
and intrusions) through Short Messaging System or Bluetooth technology.
Upon receiving a notification on a mobile phone the user initiates
appropriate action which will be implemented by the system [21]. The use
of Bluetooth for communication reduces costs since communication via
Bluetooth is not charged. However, the system is inefficient in
circumstances that require high real-time data transfer. It also does not
operate in various modes once it is started off, leaving the user with little
room to make some options on the usage level of the system. Moreover, the

IJCSBI.ORG
system does not prioritise electricity usage on different household
appliances based on the available power.
[22] developed a Smart Power Saving System in smart homes for

controlling appliances with the aim of saving power. The system comprises
two modules namely fingerprint electronic door-locking and electricity
saving. It uses GSM for interaction between the microcontroller and the
phone. A user scans their fingerprint on the door-lock and if it matches, the
electricity saving module will be turned on. The electricity saving sub-unit
controls household electric devices in the home in response to the relative
conditions from different sensors installed in the room. Fan and lights are
switched on/off in response to the temperature and light intensity inside the
home [22]. However, the module had a narrow scope in terms of monitoring
and controlling electricity usage. This is evidenced by the system’s target on
small household appliances specifically fan and lights whilst larger
appliances with higher consumption were excluded such as stoves and
fridges. The biometric module in the system added unnecessary costs as far
as the power saving was concerned. The biometric module was more into
security than power saving.
Artificial Intelligent-based systems have also been proposed for power
usage optimization. These learn about the behavior of an inhabitant in a
smart house to self-adjust the system so that it can be independent and easy
to personalize [23]. Of late, several of such Artificial Intelligent methods for
recognizing user activities using supervised learning in a smart home have
been published [24]. The main disadvantage of such systems though is that
they need prior labeled data for training the algorithms. Manually
representing human behavior data in line with event sensor readings takes a
lot of time and is monotonous and makes the system less scalable [25].
Furthermore, it is highly impractical in reality that all further inhabitant
activities will be similar with training data, thus making such systems more
suitable only for the homes for which they were designed [26]. [6]
implemented a home automation system using Arduino and Android, but
their focus was more on smart homes and the comfort associated with them
rather than power saving.
3. METHODOLOGY
The design science research methodology (Improvement Research) was
adopted for this research. The approach focuses on creation, invention or
design of some new artifacts, while deriving or obtaining suggestions to
solving the problem from current knowledge or theory base for the problem
domain [27]. Figure 1 shows the architecture of the proposed system.

IJCSBI.ORG
Sensors
Figure 1. System Architecture
Current sensors are connected to AC power source to which home
appliances are connected. The current sensors communicate with Arduino
which in turn is interfaced with the GSM module. A user can interact with
the system through an Android application. For example, if an appliance is
turned on, say a fan, a current sensor will send information to the Arduino
which will communicate with the GSM module, and a user will be notified
either via SMS or Bluetooth in the form of a system status. A user can also
turn on or off appliances remotely via an Android application through the
GSM module, Arduino and a relay action will be sent to the AC power
source. This means a user can turn on or off any appliance in the home from
anywhere. When power is restored after a power cut, the user will be
automatically notified and shown all appliances that will be on at that time
so that the user can decide which ones to turn off, thus eliminating
unnecessary power wastages. Temperature, motion and light sensors will
provide values to the Arduino, and depending on the rules set and values
read it will send appropriate relay actions to the power source. The system
also generates graphical electricity consumption reports, showing which
appliances consumed how much electricity per given period.

IJCSBI.ORG
START
Initialise sensors,
GSM and LCD
Delay for 10
seconds
Read motion sensor

status
Motion Turn OFF lights and

NO
detected? fan if ON
NO
NO
Read light
Read temperature
intensity sensor
sensor value
value
Temperature Light sensor

sensor value<set value> set
point? value?
YES YES
Turn off fan via relay Turn off light via relay
Notify user (SMS) Notify user (SMS)
Delay for
5 minutes
Figure 2. System Flow Chart
The motion sensor is used to check whether there is anyone in the house. If
there is no one yet lights and/ or fan is on, the system will automatically turn
them off. The assumption is someone might have forgotten to switch them
off before leaving the room. The system will check again whether there is
anyone in the room after ten seconds. This ten seconds delay can be set to
another value as determined by the user in line with their requirements. If
motion is detected, the system checks whether there is enough light intensity
and heat as determined by the user. If light intensity is too low, lights will
automatically be turned on. Conversely, if light intensity is too high, lights
will be automatically turned off. The fan will also be turned on if

IJCSBI.ORG
temperature is higher than a user set value and will be turned off if
temperature rises to a maximum desirable value. When 5 minutes have
elapsed, the system will check again whether there is anyone in the room by
reading a motion sensor status. This iterates as long as the system is up and
running.
The Algorithm
START
Let:
MaxTemp be the maximum room temperature before fan automatically
turns on
MinTemp be the minimum room temperature before fan automatically t
urns off
MaxLight be the maximum room light intensity before lights
automatically turn off
MinLight be the minimum light intensity before lights automatically
turn on
Note: User sets custom values to MaxTemp, MinTemp, MaxLight and
MinLight according to their preferences.
Steps:
I. Initialise sensors, GSM and LCD
II. Read motion sensor status
a. If motion is detected
i. Read temperature sensor value
1. If temperature sensor value < MinTemp Then
a. Turn off fan and notify user on mobile
phone
b. Delay for 5 minutes and goto to step I
2. Else if temperature sensor value> MaxTemp
Then
a. Turn on fan and notify user
3. Else delay for 6 seconds and goto step II.a.i.
4. End if
ii. Read light intensity sensor value
1. If light intensity > MaxLight Then
a. Turn off lights
2. Else if light intensity < MinLight Then
a. Turn on lights and notify user

IJCSBI.ORG
3. Else delay for 6 seconds and goto step I.a.ii.
b. Else
i. Delay for 5 minutes
ii. Goto step II
End If
III. If stove is turned ON Then
a. If fridge is ON Then
i. Automatically turn OFF fridge and notify user
ii. End If
b. End If
IV. If user turns stove OFF Then
a. If fridge is OFF Then
i. Automatically turn ON the fridge and notify user
ii. End If
b. End If
The system was implemented using the Java Programming language, which
was used to link the user interface and the SQLite database. In addition to
Java code, XML was used to create the interfaces for the android
application. Eclipse Indigo IDE was used to implement the application. The
Android SDK and ADT were also used. The researchers created, compiled,
debugged and deployed the android application from the Eclipse IDE using
the android ADT. The Android SDK was integrated into the Eclipse IDE to
help create and test the system during different iterations of the application.
SQLite was used for the database.
The following hardware components are required in the development of the
prototype and testing of the prototype: Arduino UNO (R3), GSM Module
with an unlocked SIM card, 4 Channel 5 volt Relay, Connecting wires,
Bread board, 16x2 LCD, Power supply, An Android mobile phone for
hosting the user application, Sensors (PIR motion sensor, ACS712 current
sensor, LM35 temperature sensor and LDR light intensity sensor), and
Resistors. Android Studio, Arduino Development Tool, Eclipse IDE and
Proteus must also be installed on the development computer.

IJCSBI.ORG
4. RESULTS
The system was evaluated in terms of its ability to optimise power usage by
domestic appliances. While a prototype was developed and tested using a
fan, light bulb, stove and water heater as the household appliances, its
effectiveness in terms of power consumption optimisation was measured for
the light bulb only. The researchers calculated light bulb power
consumption over 24 hours for best and worst cases. The researchers then
ran a 24 hours long experiment using the same light bulb on a prototype of
the proposed system. The results of these experiments are shown in Table 1.
The results indicate a significant drop in power usage when using the
system being proposed herein. Monthly figures are derived from the average
daily figures obtained from the experiments. The light bulb used was a
230V, 100W bulb which consumes 0.1 kW per hour.
Table 1. Power Consumption Comparison for a light bulb before and after installation
of the system
Before Installation After

Installation
Worst Best Optimal Case
Case Case
Appliance: Average hourly 0.1 0.1 0.1
light bulb consumption (kW)
Maximum total 24 18 12
consumption time
(Hours)
Maximum total 2.4 1.8 1.2
consumption (kW)
The worst case scenario is when an appliance remains on for the whole day
and night. Given that the bulb used consumed 0.1kW per hour, if left on for
24 hours it will consume 2.4kW. This worst case scenario is only possible if
no power optimisation system is implemented. In this experiment, the
researchers defined the best case as the case when consumption time is at
least 12 hours but less than 24 hours per day. Taking the lower bound of 12
and upper bound of 24 hours per day and calculating the average of the two,
it gives 18 hours as the best case scenario’s hours when the light will be on
per day. The assumption is that the user will be turning the lights on and off
when necessary. For 18 hours at a consumption of 0.1 kW per hour, the light
bulb will consume 1.8kW per day. The optimal case is was when the power

IJCSBI.ORG
optimisation system was used. Under this case, lights were only on for 12
hours per day, when it was dark, and hence consumed 1.2kW.
Table 2. Comparisons of Power Consumption Costs
Before Installation After

Installation
Worst Best Optimal Case
Case Case
Appliance: Average hourly 0.1 0.1 0.1
light bulb consumption (kW)
Maximum total 24 18 12
consumption time
(Hours)
Maximum total 2.4 1.8 1.2
consumption (kW)
Cost per kW per hour 0.09 0.09 0.09
(US$)
Estimated total cost per 0.216 0.162 0.108
day (US $)
Estimated total cost per 6.48 4.86 3.24
month (US $)
The researchers used Equation 1 for calculating electricity cost savings as a

percentage:
𝐸𝐶𝐵𝐼 −𝐸𝐶𝐴𝐼 ∗100
𝐸𝐶𝑆𝑃 = ... .… Equation 1
𝐸𝐶𝐵𝐼
Where
ECSP = Electricity Cost Savings as a percentage,
ECBI = Electricity Cost Before Installation of the system,
ECAI = Electricity Cost After Installation of the system
Cost savings were calculated for both the worst and best case scenarios
using Equation 1. For the worst case scenario:
𝐸𝐶𝐵𝐼 − 𝐸𝐶𝐴𝐼 ∗ 100
𝐸𝐶𝑆𝑃 =
𝐸𝐶𝐵𝐼

IJCSBI.ORG
$(6.48 − 3.24) ∗ 100
=
$6.48
∴ 𝑬𝑪𝑺𝑷 = 𝟓𝟎%
For the best case scenario:

𝐸𝐶𝐵𝐼 − 𝐸𝐶𝐴𝐼 ∗ 100
𝐸𝐶𝑆𝑃 =
𝐸𝐶𝐵𝐼
$ 4.86 − 3.24 ∗ 100
=
$4.86
∴ 𝑬𝑪𝑺𝑷 = 𝟑𝟑 𝟏 𝟑 %
Cost of electricity is directly proportional to usage, hence reducing
electricity consumption results in reduced cost of electricity to domestic
electricity consumers. In terms of percentages, reduction in cost is equal to
reduction in the amount of power consumed. The results indicate that
implementing the household power optimisation and monitoring system
resulted in cost saving of 33.3 per cent and 50 per cent for the best and
worst case scenarios respectively. Consequently, it means the amount of
power consumed was reduced by the same margins. The system resulted in
optimal power usage and thus reduced demand for electricity. Apart from
power usage optimisation, the system improves comfort levels for users as
they remotely monitor and control their household devices. The ability of
users to monitor and control household electric devices from a distance is
also useful for people living with disabilities as they can control and
monitor appliances in the home without having to physically move around
to power switches which are usually mounted on different points on the
walls of houses.
A number of authors who have been engaged in making smart homes
systems concentrated more on improving the comfort for inhabitants than
electricity saving. [28], [29] and [30] have focused more on automation with
little inclination towards power saving, hence they did not state how much
could be saved by implementing their systems. While some of the systems
resulted in electricity savings the researchers did not quantify the electricity
savings; thus providing no comparison basis. [14] claim that their system
reduced electricity consumption by domestic heating but did not state by
how much.
5. CONCLUSIONS
The Household Power Optimisation and Monitoring System proposed
herein focused mainly on reducing the amount of electricity consumed by
households and hence reducing stress on the national power grid. The
results of the system indicate that the system can reduce power consumption
in households by up to 50 per cent. This 50 per cent reduction in electricity

IJCSBI.ORG
consumed translates to 50 per cent savings in electricity costs to households.
It is important to save electricity since electrical power is scarce in
developing countries like Zimbabwe [11]. Saving electricity in households
increases the amount of electricity available for industrial use, which in turn
increases employment creation. [11] claims that every single occupation in
the manufacturing sector generates more than two million occupations in
other sectors of the economy, hence it is imperative to make sure that there
is enough electricity for the manufacturing industry. The system also has
other benefits of convenience and comfort since users can remotely
manipulate appliances on their phones. This feature makes this system an
inclusive solution as it also helps people living with disabilities to
manipulate appliances on their own without having to move around to
different power switch points around the home to power on or off
appliances. However, the system could be improved by incorporating voice
commands to control appliances. It could also be improved by adding a
functionality of predicting future consumption of an appliance based on past
and present consumption patterns. The system is designed on the
assumptions that supply of electricity is always less than demand; the users
are not using the available electricity optimally and all policies pertaining
electrical usage are held constant. The performance of this system depends
on the performance of the sensors. Moreover, remote manipulation of
electric gadgets will depend on the availability of network, hence remote
manipulation and monitoring may not work if there is no network coverage,
unless the user is within the Bluetooth range. The focus of the study is to
optimise power usage in households only.
6. REFERENCES
[1] Nyasha Kaseke, "Journal of Business Management & Social Sciences Research,"
Emergence Of Electricity Crisis In Zimbabwe, Reform Response And Cost
Implications, vol. 2, no. 10, pp. 1-16, October 2013.
[2] Tichaona Chifamba. (2011) Commercial Farmers Union of Zimbabwe Web site.
[Online]. http://www.cfuzim.org/index.php/newspaper-articles-2/zesa/2516-
continuing-power-shortages-cripple-zimbabwe-economy
[3] ZESA. (2013, January) ZESA Holdings Web site. [Online].
http://www.zesa.co.zw/index.php/component/k2/item/17-why-do-we-have-load-
shedding?
[4] Tom Chuma. (2012, Oct.) A case study of zimbabwe. [Online].
https://leadpanafricansession.files.wordpress.com/2012/11/zimbabwe-case-
study1.doc
[5] Abel Oyuke, Halley Peter Penar, and Brian Howard, "Off-grid or 'off-on':Lack of
access, unreliable electricity supply still plague majority of Africans," Afrobarometer
Dispatch, vol. 6, no. 75, pp. 1-26, March 2016.

IJCSBI.ORG
[6] S.E. Mangwengwende, "Increasing Electricity Access While Ensuring Financial
Viabilityy: A Perspective form the African Electricity Industry," in Global Network
on Energy For Suatinable Development (GNESD), Nairobi, 2005.
[7] Central Statistics Office, "Power Statistics in Zimbabwe," Power Statistics in
Zimbabwe, 2008.
[8] Giuseppe Anastasi, Francesco Corucci, and Francesco Marcelloni, "An Intelligent
System for Electrical Energy Management in Buildings," in 11th International
Conference on Intelligent Systems Design and Applications, 2012.
[9] Allan Shirichena. (2014) http://www.technomag.co.zw. [Online].
http://www.technomag.co.zw/2014/11/04/zimbabwes-energy-crisis-demand-supply-
imbalance/#stash.
[10] Phillimon Mhlanga. (2014) Financial Gazzette. [Online].
http://www.financialgazette.co.zw/zim-faces-unprecedented-power-crisis/
[11] The United Nations. (2016, August) The United Nations Web site. [Online].
http://www.un.org/sustainabledevelopment/wp-content/uploads/2016/08/9_Why-it-
Matters_Goal-9_Industry_1p.pdf
[12] Jan M Rabaey, Anantha Chandrakasan, and Borivoje Nikolic, Digital Integrated
Circuits, 2nd ed., 2002.
[13] University of Kentucky. (2014) KENTUCKY POULTRY ENERGY EFFICIENCY
PROJECT. [Online].
http://www2.ca.uky.edu/poultryprofitability/Production_manual/Chapter11_Monitori
ng_and_control_systems/Chapter11_overview_of_systems.html
[14] J. Villar, E. de la Cal, and J Sedano, "A fuzzy logic based efficient energy saving
approach for domestic heating systems.," Integrated Computer-Aided Engineering,
vol. 15, pp. 1-9, 2009.
[15] Y. He, "Energy saving of central air-conditioning and control system:," Energy
saving of central air-conditioning and control system: Case study: Nanchang
Hongkelong Supermarket., vol. 1, p. 12, 2010.
[16] M. Nowak and A. Urbaniak, "Utilization of intelligent control algorithms for thermal
comfort optimization and energy saving.," in 2011 12th IEEE Control Conference,
2011.
[17] E. Inji, I. Attia, and P Hamdy, "Energy Saving Through Smart Home.," Energy
Saving Through Smart Home., vol. 2, pp. 22-34, 2011.
[18] M. Jahn, "The Energy Aware Smart Home.," The Energy Aware Smart Home., pp. 1-
9, 2010.
[19] Borlase, Smart Grids: Infrastructure, Technology and Solutions, CRC Press, Ed.:
Taylor and Francis Group, 2012.
[20] S.P.S. Gill, "Smart Power Monitoring System Using," Sixth International Conference
on Sensing Technology, p. 1, 2012.
[21] Vini Madan, "GSM-Bluetooth based Remote Monitoring and Control System with
Automatic Light Controller," International Journal of Computer Applications, p. 1,
2012.

IJCSBI.ORG
[22] M, S Madhu, M Gangadhar, and G, C Sanjaya, "ARM Based Smart Power Saving
System for Home Automation," International Journal of Computer Science and
Information Technologies, Vol. 5 (3), pp. 2910-2913, 2014.
[23] Bengio Roux, "Deep belief networks are compact," Neural Computation, vol. 22, no.
8, p. 192, 2010.
[24] F., Rokach, L., Shapira, B., Ricci, Recommender Systems Handbook. Boston, MA:
Springer US, 2011.
[25] Hyong-Euk Lee and Young-Min Kim, From Smart Home to Smart Care, G Giroux &
H Pigot (Eds), Ed.: IOS Press, 2005.
[26] D. Cook, "A method for mining and monitoring human activity patterns in home-
based health monitoring systems.," ACM Transactions on Intelligent Systems, vol. 4,
p. 9, 2013.
[27] C.S Peirce, Collected Papers of Charles Sanders Peirce, C and Weiss , P. Eds
Harshorne, Ed. Cambrigdge: Havard University Press, 1985.
[28] Poonam B. Patil, Rupali.R Patil, Swati V. Patil, and Avadhoot R. Telepatil, "Home
Automation System Using Android and Arduino Board," International Journal of
Innovative Research in Science, Engineering and Technology, vol. 5, no. 4, pp. 5076-
5082, April 2016.
[29] M.L Sharma, Kumar Sachin, and Mehta Nipuri, "SMART HOME SYSTEM USING
IOT," International Research Journal of Engineering and Technology, vol. 4, no. 11,
pp. 1108-1112, November 2017.
[30] P Sushma and M.J Roopa, "Wi-Fi Based Home Automation System Using Androis &
Arduino Platform," International Journal for Research in Applied Science and
Engineering, vol. 5, no. VI, pp. 942-945, June 2017.

Batani, J., Dzambo, S. and Magodi, I., 2017. Household Power Optimisation
and Monitoring System. International Journal of Computer Science and

Vol 17 No 2 - July-December 2017

Încărcat de

Informații document

Drepturi de autor

Formate disponibile

Partajați acest document

Partajați sau inserați document

Opțiuni de partajare

Vi se pare util acest document?

Este necorespunzător acest conținut?

Drepturi de autor:

Formate disponibile

Vol 17 No 2 - July-December 2017

Încărcat de

Drepturi de autor:

Formate disponibile

International Journal of Computer Science

and Business Informatics

ISSN: 1694-2507 (Print)

An Adaptive and Real-Time Fraud Detection Algorithm in Online Transactions .................................... 1

Agronomic Disaster Management using Artificial Intelligence - A Case Study ..................................... 13

Household Power Optimisation and Monitoring System ................................................................... 23

An Adaptive and Real-Time Fraud

ISSN: 1694-2108 | Vol. 17, No. 2. JULY-DECEMBER 2017 1

ISSN: 1694-2108 | Vol. 17, No. 2. JULY-DECEMBER 2017 2

ISSN: 1694-2108 | Vol. 17, No. 2. JULY-DECEMBER 2017 3

ISSN: 1694-2108 | Vol. 17, No. 2. JULY-DECEMBER 2017 4

The continuous updating of the cardholder‟s financial profile through ANNs

Figure 1. System flow chart

ISSN: 1694-2108 | Vol. 17, No. 2. JULY-DECEMBER 2017 5

3.3 Algorithm for the Proposed System

ISSN: 1694-2108 | Vol. 17, No. 2. JULY-DECEMBER 2017 6

Neural Network Algorithm

ISSN: 1694-2108 | Vol. 17, No. 2. JULY-DECEMBER 2017 7

ISSN: 1694-2108 | Vol. 17, No. 2. JULY-DECEMBER 2017 8

Table 2. Testing Results

ISSN: 1694-2108 | Vol. 17, No. 2. JULY-DECEMBER 2017 9

Sensitivity is regarded as the most important measure of the effectiveness of

In comparison to other systems in existence, the proposed algorithm is

ISSN: 1694-2108 | Vol. 17, No. 2. JULY-DECEMBER 2017 10

ISSN: 1694-2108 | Vol. 17, No. 2. JULY-DECEMBER 2017 11

This paper may be cited as:

ISSN: 1694-2108 | Vol. 17, No. 2. JULY-DECEMBER 2017 12

Agronomic Disaster Management

ISSN: 1694-2108 | Vol. 17, No. 2. JULY-DECEMBER 2017 13

usually make use of rough set theory, neural network, evolutionary

ISSN: 1694-2108 | Vol. 17, No. 2. JULY-DECEMBER 2017 14

In [6], a neuro-fuzzy model was developed to predict the monthly

3. CASE STUDY REGION

4. MATERIALS AND METHODS

ISSN: 1694-2108 | Vol. 17, No. 2. JULY-DECEMBER 2017 15

‘no’ denotes no precipitation (rainfall) otherwise if Pp = ‘yes’ indicates

Table 1. Observatory weather Dataset

5. AGRONOMIC DISASTER MANAGEMENT MODEL

This decision support system modelled for agronomic disaster management

ISSN: 1694-2108 | Vol. 17, No. 2. JULY-DECEMBER 2017 16

The pseudocode of the multi-layered back-propagation algorithm is as

The proposed system is developed and implemented using Microsoft

ISSN: 1694-2108 | Vol. 17, No. 2. JULY-DECEMBER 2017 17

Task relevant input

(1) BP- Perceptron model

Obtained error ≤ Desired

ISSN: 1694-2108 | Vol. 17, No. 2. JULY-DECEMBER 2017 18

any classification model utilizing artificial intelligence exhibits better

Table 2. Performance evaluation of proposed system for adaptable learning rate

Table 3. Performance evaluation of Data-driven AMD vs existing models

Exhaustive-forward Random-forest Weka-tool 81.07 %

PSO Bayesian-network Weka-tool 89.75 %

ISSN: 1694-2108 | Vol. 17, No. 2. JULY-DECEMBER 2017 19

Attained Prediction Accuracy (%)

Figure 2. Precipitation prediction accuracy optimization report

ISSN: 1694-2108 | Vol. 17, No. 2. JULY-DECEMBER 2017 20

ISSN: 1694-2108 | Vol. 17, No. 2. JULY-DECEMBER 2017 21

This paper may be cited as:

ISSN: 1694-2108 | Vol. 17, No. 2. JULY-DECEMBER 2017 22

Household Power Optimisation and

ISSN: 1694-2108 | Vol. 17, No. 2. JULY-DECEMBER 2017 23

ISSN: 1694-2108 | Vol. 17, No. 2. JULY-DECEMBER 2017 24

ISSN: 1694-2108 | Vol. 17, No. 2. JULY-DECEMBER 2017 25

[8] designed an automated power management system called the