Safety App: Crime Prediction Using GIS
Sourab Banka*, Sean D’cruz*, Atharva Deshmukh*, Sana Shaikh*
Department of Computer Engineering
Don Bosco Institute of Technology, Mumbai, India.
Abstract
Crime in Mumbai has always been a concern for
both the government authorities as well as the
citizens of Mumbai. With the increase in the
population, the crime rate has also been on an
increase which in turn increases the risk factor
for the safety of the citizens of Mumbai. To deal
with this, the citizens should not only rely on
the government authorities to cater the needs
but at the same time, the citizens should also
understand and adapt with the applications of
technology in this very era of technology, to help
provide better safety for all the citizens residing
in the dynamic city of Mumbai. Safety while
traveling can be considered as the major need
in Mumbai, as Mumbai is regarded as the city
that never sleeps. There have been several at-
tempts made by the government authorities as
well as individuals to help provide a model for
better safety while traveling but still it has never
been completely automated. This research pro-
poses a model that works in synergy with vari-
ous steps that include crime data collection, and
then analysis on that data to predict the crime
hotspots which are specifically targeted to help
citizens to distinguish between safe and unsafe
areas while traveling.
Keywords— Crime Prediction, Safety Route,
Web Scraping, Crowd-sourcing, Safety Android Ap-
plication
Introduction
In contrary to common belief, crime isn’t continu-
ally focused in poor areas, with moneyed elements of
the town affected too, therefore residents and guests
would like info to avoid additional dangerous ele-
ments of the town. the concept is to form a reli-
able source of data instead of suppose episodic news
reports that contribute to a way of hysteria. Our
goal is to form what ar already in public offered
statistics accessible and unjust for citizens. Unwary
citizens can make easy targets because they stand
out in a crowd, these citizens are usually unused to
their surroundings, or are generally carrying money,
credit cards and valuables like mobile phones, cam-
eras. Due to this, the risk of being mugged or robbed
is generally high while traveling[17]. In fact, 72% of
the criminal activities that take place in Mumbai
are due to avoiding the basic signs which could have
been prevented if there existed a system that could
help identify and distinguish between the safe and
unsafe areas in Mumbai.[11] This research model ob-
serves and realizes after studying several criminal
cases that Mumbai has encountered, that citizens
usually ignore the fact to be safe when it comes to
reaching a destination on time or in haste, and keep-
ing the same thing in mind our group aims to build
an application that helps citizens to distinguish and
predict between safe and unsafe areas of Mumbai
while traveling.
The analysis is achieved using multiple sources in-
cluding the web scrapping, official FIR reports and
user ratings for every specific area on which ma-
chine learning is applied and then created a Training
model based on the 240 official FIR reports and ad-
ditional user responses to help achieve an accurate
safety index.
Literature Survey
Nowadays, millions ofcitizensgo through the streets
of Mumbai town, takingsome specific routes that
area unit planned by victimisation either public or
non-public transportation or perhaps walking. Al-
though there area unit well-known routes that voters
most frequently regard their traveling, new routes
(probably unsafe) could be experimented especial-
lyfor newcomers. particularly, safetyis a important
characteristic that ought to be thought of for these
mobile applications, especiallyin overcrowded giant
cities. Mobile applications for routing and plan-
ningin town environments are more and more turn-
ing into essential for rising urban areas. Thus, this
means the looks of the nextgeneration of mobile
info systems, during which recommendations area
unit centered on call makingin order to adequately-
support the growth of bigcities. Applications like
CrowdPlanner [1] and DroidOppPathFinder [2] area
unit self-addressed to come up with crowdsensed
route recommendation systems, that request from
users to guage candidate routes counseled by totally
different sources and ways, for determiningthe best
route supported the feedback of these users. The
routesgenerated bythese applications area unit eval-
uated by people; in our case the assessment is cre-
ated by a clump formula and bythe crime occur-
rence. Recently, an outsized vary of experimental
applications, that take into consideration the social
network and crowdsensed information, were devel-
oped. In [4], a spatio- and crowd-based routingsys-
tem is projected so as to boost the recommendation
quality; but, it’s solely supported volunteered info
and therefore the enrichment of the crime informa-
tion sources is not considered
Firoz Ahmad, Md Meraj Uddin, Laxmi
Goparaju[3] has analyzed the crime using a
vector file of geographic information system for
India country and different states of India country.
Using this vector file they analyze crime from the
year 2001 to 2015 to learn the temporal pattern
of crime happening in India. They shortlisted
crimes such as Murder, dacoity, robbery, burglary,
theft, and riots. The analysis was done using some
proprietary as well as free software. Vector data
was downloaded from the open-source platform
and Crimes data set was available on National
Crime Records Bureau, Ministry of Home Affairs,
India. They downloaded the crime data for 10
different cities that have high crime rates. Through
this, we come to know Geo graphics information is
available on the DIVA-GIS platform[22]. Historical
crime data and its location were mapped together
and also analyzed to come to learn the pattern
of criminal behavior near such as the location of
the police station, road network, shopping malls,
buildings, recreational center.[14]
Researcher, market analyzer or academicians
gather data from different websites for their better
improvement. Copping of data on the website to
user local storage in forbidden by most of the web-
site authority. So that the user want to manually
coping the data from website to local computer
file storage. But such a task is very exhausting
and time consuming. Due to such limitation web
internet scraping techniques area unit introduces.
By exploitation net scraping techniques user will
extract info obtainable on multiple web site into
a single database or spreadsheets. So data can be
easily visualize and analyse for further use. Web
scraping technique is a sub-discipline of web mining
technology.
Proposed Architecture
Data from numerous open sources like news, open
Data, or through user input or web scraping knowl-
edge from the web site are going to be keep within
the information Server. At that time feature choice
are going to be done on knowledge accessible within
the information. Once feature selection is finished,
knowledge cleansing are going to be performed on
knowledge to get rid of vociferous knowledge and
chemical agent. Knowledge having great circle and
latitude are going to be premeditated on the QGIS
server. Finally, the Prediction algorithmic rule can
work on the given knowledge that is plotted within
the MAP.[10] The output are going to be a map
accessible for the user on android Application and
additionally for Agencies it’ll be accessible within
the sort of an internet site with completely different
visual image of knowledge options. This Application
can use advanced machine learning to predict crime
rates within the city’s neighborhoods at completely
different times of the day and night, the applying
are going to be ready to predict new patterns of
crime within the space with the assistance of incom-
ing new crime data set. The app are going to be
ready to assess the danger of crime in several ele-
ments of the town by finding patterns in incidents,
locations and time, using advanced algorithms. It
will this by process GPS coordinates furthermore as
a particular time and date.
The app filters these sectors by their magnitude to
make separate models. We have a tendency to then
weight the events by sector population and adjacent
sectors and periods before ultimately manufactur-
ing the world score. The score is that the relative
likelihood of against the law in every sector and at
completely different times, and is summarised on a
scale of 1 to 10, wherever one is low risk and 10 is
high risk, in many areas of the town. The ultimate
output can turn out a simplified and color-coded risk
rating that users of the app will see on a map. The
app shows 3 completely different classes of against
the law risk, together with bad crimes like assault,
rape, etc and theft mistreatment arms; medium risk
crimes, like stealing, etc.
Figure 1: System Architecture
The safe route will be obtained by the difference
of the Dijkstra formula, within which the nodes
within the network square measure assigned as a
mean weight that was obtained from the number of
crimes for a particular purpose or geographical area.
The values which will be generated from the math-
ematician formula mirror the chance of getting or
not an occurrence like “theft/crime.” etc. Or we can
use a additional advanced formula for locating the
shortest path are Highway/Route Hierarchies, A*
Path Finding formula.[15] The design approach is
to retrieving data from various sources that are re-
lated to crime events, taking into account attributes
that define the time and location when an event oc-
curs. These data are analyzed and integrated with
crime data from official sources (government insti-
tutions). The integration process is automatically
performed by using descriptions and spatiotemporal
attributes of data.[6] The categorization and clas-
sifying are carried out analyzing the spatiotempo-
ral attributes and the description of words that ap-
pear in the data available through various sources
and/or official database records. Thus all the cat-
egorizedand classified knowledge ar used as input
parameters forthe safe routing algorithmic program.
the ultimate result’s a saferoute that doesn’t cross
or contain points with highcrime rates.
Techniques
1. CrowdSourcing: The survey results clearly
pointed to us that Lighting, Walk Path, Visibility,
Security and Transport were the important 5 pa-
rameters. the particular area is rated based on the
above 5 considered parameters and each one of the
parameters can be rated in the index of 1-5, based
on what scale the parameter that is being considered
is true.
Thus, finally on the basis of rating, it will tell us
that the area on the right side has a much higher
safety index than the area on its left thus making it
safer and less crime-prone.
Steps for Uploading images and data in FIREBASE
database
I. Set Firebase Permissions: To allow your app
access to Firebase Storage, you need to set up per-
missions in the Firebase console. From your console,
click on Storage, and then click on Rules.
II. Create the Application: Open up Android Stu-
dio, and create a new project
III. Set Up the MainActivity Layout: The appli-
cation will need one activity layout. Two buttons
will be needed—one to select an image from your
device, and the other to upload the selected image.
IV. Get MainActivity Up: Navigate to your Main-
Activity, and start by declaring fields. These fields
will be used to initialize your views (the buttons and
ImageView), as well as the URI indicating where the
image will be picked from. Add this to your main
activity, above the onCreate method.
V. Uploading the File to Firebase: folder gets cre-
ated automatically when the image is uploaded. Lis-
teners are also added, with toast messages. These
messages get displayed depending on the state of the
upload.
VI. Set Permission in the App: Finally, you need
to request permission that your application will
make use of. Without this, users of your application
will not be able to browse their device gallery and
connect to the internet with your application. Do-
ing this is easy—simply paste the following in your
AndroidManifest file
VII. Testing the application: User will upload the
images and rate the parameter on scale from 1 to 5.
Firebase provides developers with several advan-
tages, and file transfer with storage is one in all
them. Uploading pictures from your android ap-
plication needs you to figure with Activities and In-
tents.
The user of the application uploads images of a par-
ticular area as should below in figure(2).
2. Web Scraping: Web scraping is a process of au-
tomating the extraction of data in an efficient and
fast way. With the help of web scraping, you can
extract data from any website, no matter how large
is the data, on your computer. Scraping from dif-
ferent news websites related to crime will be done.
Important data from the news will be added to the
database.
Following are the steps that actually takes place
when the program runs:
I.Import libraries
II.Set the URL you want to web scrape from
III.Connect to the URL
IV.Parse HTML and save to BeautifulSoup object
V.To download the whole data set, we do a for loop
through all <a> tags
VI.Pause the code for a sec(To prevent ourselves
from getting banned)
Web scraping is a process of automating the ex-
traction of data in an efficient and fast way. With
the help of web scraping, you can extract data from
any website, no matter how large is the data, on
your computer. Moreover, websites may have data
that you cannot copy and paste. Web scraping can
help us extract any kind of data that you want.[4]
Following are the steps of performing web scrap-
ping :
A. Open the website on which you want to perform
web scrapping and based on what kind of data you
want to extract start inspecting the particular el-
ements from the console and try to figure out the
tags and attributes that are of relevance to your web
scrapping process.[4]
B. Then Start by importing the python libraries
import requests
import urllib.request
import time
from bs4 import BeautifulSoup
The most important library here is the Beautiful-
Soup.
Beautiful Soup is a Python library for pulling data
out of HTML and XML files. It works with your
favorite parser to provide idiomatic ways of navi-
gating, searching, and modifying the parse tree. It
commonly saves programmers hours or days of work.
To parse the html with BeautifulSoup so that we
can work with a nicer, nested BeautifulSoup data
structure.
We use the method .findAll to locate all of our
required tags. We should include time.sleep(1) line
of code so that we can pause our code for a second so
that we are not spamming the website with requests.
This helps us avoid getting flagged as a spammer.
From the news article from TimesofIndia paper
 Figure 2: Web Scrapping Output

has scraped the important data we are required for

the project. Data can be places, type of crime,
day, age of the victim, etc which is available in any
news article of crime-related.[23] These data were
extracted from news and appended into a csv. in
figure (2). You can observe the same data extracted
from the news.
3. Official Reports: Under the public open-source
portal, Maharashtra police provide the facility to
view published FIR online. FIR can be downloaded
in a variety of formats like CSV, JSON, PDF, etc.
CID Branch also provides the weekly and monthly
statistical reports of each District[17]. These reports
can be useful to analyze the crime on a monthly or
weekly basis[19]. Figure 4: Crime Density plot of Murders in
Mumbai

Figure 3: Maharashtra Police - Services for Cit-

izen Portal

Maharashtra Police has provided a citizen portal to

access the FIR published on Internet figure (3). Por-
tal allows us to view FIR published in any 3 months
range of specifically unit areas. This portal gives
latest records as per the search parameters specified
in search query. Below is snapshot of available FIR
to download.
At once, you can download only one FIR. FIR is
available in multiple formats such as CSV, JSON,
PDF, etc. FIR has some important fields which can
be used in our project system. Some of the impor-
tant parameters which we can extract are Age of
victim, Place,Date, Time, etc. Once the parame-
ters are extracted we can insert into main Database
which is Integrated with other Data source.
The approach towards building an app is based on
Figure 5: Crime Density plot of Thefts in Mum-
three aspects which are Official Reports, Web scrap-
ping and Crowdsourcing. We will train our model
bai
initially using Official Reports and then build it up
in the sequential phases using Web scrapping and
Crowdsourcing, thus we are trying to build an sys-
tem that works on all the aspects and integrates all
technologies into one, which is what makes it dif-
ferent from some similar projects which have been
done earlier.

Implementation
This mobile application will be using Django as
backend, its flexible, stable and customizable. The
Backend API will have a simple model for Location,
Figure 7: Safety Index Analysis
Rating and Prediction object, which will have its
respective serializer class created using Django Rest
framework. The Backend API will also have respec-
Finally, we have come to a conclusion based on
tive API functions in views.py which will server re-
the research and the analysis we conducted is that
quests at end point. The Frontend Mobile Applica-
cities like Andheri, juhu and kurla are comparatively
tion is created using React Native. It works with
not very safe for the users.
Expo SDK and apps so testing the application is
convenient on mobile devices as well as simulator.
The Application fetches data from the server and
renders it to the view. In Short, this application
Result Analysis
will only be developed for Android, React Native is The Graph in figure (7) shows the plotting of Crime
powerful framework for building both iOS and An- committed at different time intervals in Mumbai
droid apps. city. Thus, from the graph analysis it is seen that
between 12AM to 12PM most of the crimes are com-
mitted.

Figure 8: Estimate of Crimes at Various Interval

The below mentioned graph depicts the crime

rates at different parts of the city ranging from low
to high based on the number of thefts, robberies
and/or murders and also based on the 245 responses
we accumulated during the initial data collection.
It can very clearly observed that cities like Andheri
and Bandra have a high rate of number of thefts,
similarly andheri, bandra, dadar have a high rate
of number of robberies, and also andheri, kurla and
mulund have a high rate of number of murders.
The Graph in figure (8) shows the plotting of var-
ious crime events like Murder, Theft and Robbery
in several cities of Mumbai. It is clearly visible that
Figure 6: Model Predication Output on App In- Theft is the most committed crime in all the cities
terface and Murder being the least frequent one

The Graph in figure (10) shows the plotting of

Safety Index Value in several cities of Mumbai. The
lower the Safety value the more crime prone the lo-
cation. Thus, from the graph analysis it is obvious
that Andheri, Kurla and Bandra are unsafe because
it has the least safety index values.
Conclusions
In this paper, the complete analysis, classification
and prediction of the Mumbai City is done us-
ing crowdsourcing information with the information

Figure 9: Crimes in Different Cities
provided from social network and also the official
crime reports which in this case is FIR.The spa-
tiotemporal analysis can confirm the situation wher-
ever the crime event occurs. Moreover, the confi-
dence level of a location is going to be outlined and
it’ll be used as a parameter for computing the safer
route.
References
[1] H. Su, K. Zheng, J. Huang, H. Jeung, L. Chen,
and X. Zhou, “Crowdplanner : a crowd-based
route recommendation system,” inProceeding-
softhe30thIEEEInternationalConferenceonData
Engineering(ICDE’14),pp.1144–
1155,IEEE,Chicago,Ill,USA, April2014.
[2] V. Arnaboldi, M. Conti, and F. Delmastro,
“Implementation of CAMEO: a context-
aware middleware for opportunistic mo-
bile social networks,” in Proceedings of the [14] Alkesh Bharati, Dr. Sarvanguru RA.K. “Crime
IEEE International Symposium on a World
of Wireless, Mobile and Multimedia Net-
works(WoWMoM’11),pp.1–3,Lucca,Italy,June2011. and Technology (IRJET)
[3] Firoz Ahmad, Md Meraj Uddin, Laxmi
Goparaju," Role of Geospatial Technology in
Crime Mapping: A Perspective View of In-
dia",World Scientific News 88(2) (2017) 211-226
[4] M. Nagarajan, K. Gomadam, A. P. Sheth,
A. Ranabahu, R. Mutharaju, and A. Jad-
hav, “Spatiotemporal-thematic analysis of cit-
izen sensor data: challenges and experi-
ences,” in Web Information Systems Engi-
neering—WISE2009,vol.5802ofLecture Notes in
Computer Science, pp. 539–553, Springer,
Berlin, Germany,2009.
[5] S.C.M. de S Sirisuriya,2015, A Comparative
Study on Web Scraping .Proceedings of 8th In-
ternational Research Conference, KDU
[6] Félix Mata, Miguel Torres-Ruiz, Giovanni
Guzmán, Rolando Quintero, Roberto Zagal-
Flores, Marco Moreno-Ibarra, and Eduardo
Loza,"A Mobile Information System Based on
Crowd-Sensed and Official Crime Data for Find-
ing Safe Routes: A Case Study of Mexico
City",Hindawi Publishing Corporation Mobile
Information Systems Volume 2016, Article ID
8068209,
[7] Varvara Ingilevicha, Sergey Ivanov. "Crime rate
prediction in the urban environment using so-
cial factors". 7th International Young Scientist
Conference On Computational Science (2018).
[8] Hitesh Kumar Reddy, Bhavna Sainia, Ginika
Mahajana. "Crime Prediction Monitoring
Framework Based on Spatial Analysis”. Interna-
tional Conference on Computational Intelligence
and Data Science (ICCIDS 2018)
[9] ALEXANDROS BELESIOTIS, GEORGE PA-
PADAKIS, DIMITRIOS SKOUTAS . "Analyz-
ing and Predicting Spatial Crime Distribution
Using Crowdsourced and Open Data". ACM
Transactions on Spatial Algorithms and Sys-
tems, Vol. 3, No. 4, Article 12 (2018).
[10] Isha Pradhan, Katerina Potika, Magdalini Eiri-
naki and Petros Potikas. “Exploratory Data
Analysis and Crime Prediction for Smart Cities”.
International Database Engineering Applica-
tions Symposium (IDEAS 19).
[11] Aman Kumar, Nikhil Tiwari, Prakhar Gupta,
Dr.S.N.Rajan. “WOMEN CRIME PREDIC-
TION”. International Research Journal of Engi-
neering and Technology (IRJET).
[12] Seung Hwan Bang, Bongjun Ji. “Real-
time Crime Prediction Method using Criminal
Records”. Conference Paper - June 2016.
[13] Ginger Saltos, Mihaela Cocea. “An Exploration
of Crime Prediction Using Data Mining on Open
Data” Article in International Journal of Infor-
mation Technology and Decision Making · May
2017
Prediction and analysis using Machine Learning”
International Research Journal of Engineering
[15] Chao Huang, Junbo Zhang, Yu Zheng, Nitesh
Chawla. “Deepcrime: Attentive Hierarchical Re-
current Networks for Crime Prediction”. Inter-
national Journal.
[16] https://www.hindustantimes.com/mumbai-
news/mumbai-s-crime-rate-shoots-up-in-
2018/story-1sUP9o33Ga53tADqjKjhKL.html
[17] http://mahacid.com/monthly-clime-
review.html
[18] http://www.mhpolice.maharashtra.gov.in/Citizen/MH/
[19] http://mahacid.com/StateBrpg.html
[20] http://ncrb.gov.in/
[21] https://data.gov.in/
[22] https://www.diva-gis.org/
[23] https://timesofindia.indiatimes.com/city/mumbai/mumbai-
two-arrested-for-stealing-goods-from-moving-
train-coaches/articleshow/71788161.cms