See discussions, stats, and author profiles for this publication at: https://www.researchgate.

net/publication/235421872

Predication of Parkinson's disease using data mining methods: A comparative

analysis of tree, statistical and support vector machine classifiers

Article  in  Indian Journal of Medical Sciences · June 2011

DOI: 10.4103/0019-5359.107023 · Source: PubMed

CITATIONS READS

23 479

3 authors, including:

Yugal Kumar Gadadhar Sahoo

Birla Institute of Technology, Mesra Birla Institute of Technology, Mesra
45 PUBLICATIONS   305 CITATIONS    167 PUBLICATIONS   1,281 CITATIONS   

SEE PROFILE SEE PROFILE

Some of the authors of this publication are also working on these related projects:

Dengue Disease Prediction View project

Correlate the microstructre with mechanical property by using EBSD Technique View project

All content following this page was uploaded by Yugal Kumar on 22 May 2014.

The user has requested enhancement of the downloaded file.

 65

JUNE 2011

ORIGINAL ARTICLES
Clinical efficacy of subgingivally delivered 0.5% controlled release azithromycin gel
in the management of chronic periodontitis
Prashant Tyagi, Shubhra Vaish, Vidya Dodwad ............ ............ 223
Predication of parkinson's disease using data mining methods: A comparative analysis of tree,
statistical, and support vector machine classifiers
Geeta Yadav, Yugal Kumar, Gadadhar Sahoo ............ ............ 231
The effect of sildenafil on the elasticity of erythrocytes in homozygous sickle cell disease
Abigail C. Hagley, Novie O. Younger-Coleman, Andrea A. Richards,
Chukwuemeka R. Nwokocha, Dagogo J. Pepple ............ ............ 243
In-vitro evaluation and comparison of the anti-microbial potency of commercially available
oral hygiene products against Streptococcus mutans
R. Sentila, A. Gandhimathi, S. Karthika, R. Suryalakshmi, A. Michael ............ ............ 250
Dengue epidemiology in thanjavur and Trichy district, Tamilnadu - Jan 2011-Dec 2011
Chinnathambi Kalidoss Bhuvaneswari, Ramalingam Senthil Raja, Kavita Arunagiri,
Shambasivam Mohana, Karuppanan Sathiyamurthy, Kaveri Krishnasamy,
Palani Gunasekaran ............ ............ 260
 231

ORIGINAL ARTICLE

PREDICATION OF PARKINSON’S DISEASE USING DATA MINING

METHODS: A COMPARATIVE ANALYSIS OF TREE, STATISTICAL, AND
SUPPORT VECTOR MACHINE CLASSIFIERS
GEETA YADAV, YUGAL KUMAR1, GADADHAR SAHOO2

ABSTRACT

The prediction of Parkinson’s disease in early age has been challenging task among
researchers, because the symptoms of disease came into existence in middle and late
middle age. There are lots of symptoms that lead to Parkinson’s disease. But this article
focuses on the speech articulation difficulty symptoms of PD affected people and try
to formulate the model on the behalf of three data mining methods. These three data
mining methods are taken from three different domains of data mining i.e., from tree
classifier, statistical classifier, and support vector machine classifier. Performance of these
three classifiers is measured with three performance matrices i.e., accuracy, sensitivity,
and specificity. Hence, the main task of this article is tried to find out which model
identified the PD affected people more accurately.

Key words: Classifiers, decision stump, logistic regression, machine learning, Parkinson’s
and sequential minimization optimization

INTRODUCTION diseases, Parkinson’s disease is second most

Background
Presently, there are lots of neurodegenerative
diseases have been recognized such as
Alzheimer’s disease, Parkinson disease,
Arthritic disease, Dementia with Lewy bodies,
Corticobasal degeneration, Progressive
supranuclear palsy, Prion disorders, and so
on. Among all of these neurodegenerative
common disease after Alzheimer’s.[1] James
Parkinson described the core clinical feature
of the Parkinson’s disease.[2] James Parkinson
described the Parkinson’s disease as paralysis
agitans. But later on surname of James
Parkinson i.e., Parkinson’s was adopted as
disease name. Parkinson’s disease affects the
neuron cell in the brain of live organism. The
function of these neurons cells is controlling

Departments of Pharmaceutical Sciences and Access this article online

2
Information Technology, Birla Institute of Technology, Quick Response Code: Website:
Mesra, Ranchi, Jharkhand, 1Department of CSE/IT, www.indianjmedsci.org
Hindu College of Engineering, Industrial Area, Sonepat,
Haryana, India DOI:
10.4103/0019-5359.107023
Address for correspondence:
PMID:
Mr. Yugal Kumar,
VPO‑Musepur, Disst.‑Rewari Haryana, India ‑ 123 401. *****************************
E‑mail: yugalkumar.14@gmail.com

Indian Journal of Medical Sciences, Vol. 65, No. 6, June 2011

232 INDIAN JOURNAL OF MEDICAL SCIENCES
and coordinating the body movements. But in
this disease, chemical imbalance occurs in two
brain chemical of live organism i.e., dopamine
and acetylcholine. These chemicals are
responsible for controlling and coordinating
the body movements. Imbalance between
these two chemicals affects the control and
coordination of body movement. Hence, due
to this chemical imbalance, central nervous
system (CNS) of human being is also affected.
A number of environmental factors have been
implicated in PD.[3] However, aging is the most
significant risk factor for development of the
disease. It affects both sexes (more male
prevalence). [4] With the number of people
in the United States aged above 65 years
expected to double by 2030.[5] According to[6],
in India 11,747,102 number of people out of 1,
065, 070, 6072 are affected from Parkinson’s
disease. These statistics are generated with
help of extrapolations of various prevalence
or incidence rates against the populations of a
particular country or region but these statistics
are not fully reliable. William Dauer and Serge
Przedborski had classified the Parkinsonian
Syndromes i.e., Primary Parkinsonism,
Secondary Parkinsonism, Parkinson‑plus
Syndromes and Familial Neurodegenerative
Diseases. [7] Parkinson’s disease symptoms
do not occur in late middle age and vary from
individual to individual. Symptoms of Parkinson
disease begin subtly and advanced slowly.
The classical and earlier symptoms include
trembling in the hands, arms, legs and face,
rigidity, brakykinesia, and asymmetric onset.[8]
Besides these classical symptoms, some
other symptoms are also found that lead to
Parkinson’s disease. These symptoms are
micrographia (small almost unreadable writing),
decreased olfaction and postural instability,
Indian Journal of Medical Sciences, Vol. 65, No. 6, June 2011
slowing of the digestive system, constipation,
fatigue, weakness and hypotension.[9] As the
disease progresses, patient also faces some
other problems such as motor skills, speech,
gait, digestion, movement, emotion, blood
pressure, balance and posture. Recent studies
on Parkinson’s disease conclude that most of
PD patients (approximately 90%) have suffered
speech difficulties i.e., dysphonia (impaired
speech production) and dysarthria (speech
articulation difficulties). [10] These mobility
deficits are difficult to treat with drugs or
neurosurgery.[11,12]
Machine learning algorithms in disease
Machine learning algorithms have good history
in disease diagnosis and prediction. A large
number of papers have been published that
exhibited the application of machine learning
algorithm in medical field such as diagnosis
of disease, prediction of disease, survivability
and identification of disease. Initially, three
branches of machine learning came into view
i.e., symbolic learning, statistical methods
and neural networks. Symbolic learning was
described by Hunt, [13] statistical methods
described by Nilsson[14] and neural networks
by Rosenblatt.[15] Machine learning community
has developed large number of machine
learning tools that have been widely used to
obtain classification models including medical
prognostic models.[16,17] For cancer diagnosis
and research, artificial neural network and
decision tree classifiers have been used
and these methods provided remarkable
results.[18,19] Pendharker applied several data
mining methods to diagnosis patterns in
breast cancer.[20] Dursun Delen et al., applied
ANN, Decision Tree and Logistic Regression
method to predict the survivability of breast
 PREDICATION OF PARKINSON’S DISEASE
cancer patient. [21] Logistic regression and
K‑ nearest neighbor model with six different
machine learning algorithms had used to
predict the pneumonia mortality.[22] Support
vector machines (SVMs) have been used
for detection and diagnosis of wide range
of biomedical diseases such as detection
of oral cancers in optical images,[23] polyps
in CT colonography, [24] detection of micro
calcifications in mammograms, [25] and
analysis of gene expression measured via
microarrays. [26] Study of several machine
learning approaches for micro calcification
detection has shown that SVMs provide better
classification performance to other approaches
such as ANNs. [27] Bayesian networks have
been applied in biomedicine, especially
in probabilistic expert systems for clinical
diagnosis[28,29] and computational biology. [30]
Because Bayesian network has capability to
deal with biomedical data that either incomplete
or partially correct. [31] At present, machine
learning techniques are also used for detecting
and classifying tumors via X‑ray and CRT
images,[32,33] classification of malignancies from
proteomic and genomic assays.[34‑36] According
to the Pub Med statistics nearly 1,800 of
papers has been published on cancer by use
of machine learning techniques.
PREVIOUS RESEARCH
Many researchers classified the Parkinson’s
disease by several methods. Fiona O’reilly[37]
et al., determined social, psychological, and
physical aspects of a person whose partner
was affected by Parkinson’s disease and
result of his study showed that person has
slightly worst social, psychological, and
physical profiles. Samyra H.J. Keus et al.,[38]
Indian Journal of Medical Sciences, Vol. 65, No. 6, June 2011
233
had provided evidence‑based analysis of
physical therapy in Parkinson’s disease and
also developed multifaceted implementation
strategy for PD. These studies provide a
platform of physical therapy to diagnosis
Parkinson’s disease in future. A recent study on
rehabilitation of PD patient has been provided
by Khan F et al.,[39] and used SPSS version 15
to generate the results. This result includes
four parameters i.e., functional independence
measure (FIM) scores and FIM efficiency,
hospital length of stay (LOS) and discharge
destination. Hence, overall conclusion of
this study showed that most patients with
PD were in the higher functioning ANSNAP
classes (216–18) with only less than 1% in the
very disabled class (ANSNAP 219). Another
type of study that includes questionnaires to
PD affected patient had done by AGEM de
Boer et al.[40] In this study, a list of questions
were sent to patient and response of the
patient was used to determine the quality of
life in patient with Parkinson’s disease. Two
approaches were used to determine the validity
of PDQL i.e., discriminate validity of the PDQL
and convergent validity of PDQL. PDQL used
four parameters i.e., parkinsonian symptoms,
systemic symptoms, emotional functioning,
and social functioning. Result of this study
showed that Parkinson’s disease affected
patient took twice times as inflammatory
bowel disease affected patient to fill the
same questionnaires.[41] This study concludes
that PDQL is an appropriate, validate and
useful tool to determine quality of life of
patients with Parkinson’s disease in clinical
studies. Patrica Limousin et al.,[42] had provided
another method for the treatment of advanced
Parkinson’s diseases i.e., electrical stimulation
of the subthalamic nucleus. The results of
234 INDIAN JOURNAL OF MEDICAL SCIENCES
this method were tested on activities of daily
living and motor examination of the Unified
Parkinson’s Disease Rating Scale and this
method leads to decreases symptoms of
medication and levodopa dose can be reduced,
with a consequent reduction in dyskinesias.
Yunfeng Wu et al.,[43] focused on the statistical
analysis of gait rhythm in patients affected with
Parkinson’s disease. Kenneth Revett et al.,[10]
discussed rough set approach in feature
selection for Parkinson’s disease. They focused
on the impairment in voice production and used
rough set approach to identify the person with
Parkinson’s disease.
PREDICATION MODELS
In this paper, three different types of
classification methods are used i.e., decision
stump (tree classifier), logistic regression
(statistical classifier), and sequential
minimization optimization (support vector
machine).
Tree classifiers
Tree is a classifier that can be defined as
a recursive partition of the dataset. Tree
classifiers mainly consist a set of nodes in
which one of the node acts as root node;
all other nodes have exactly one incoming
and outgoing edge known as internal nodes
and rest of nodes with no outgoing edges
known as terminal nodes or leaf nodes. A tree
representation form can be easily converted
into IF‑THEN rules and used for extract
valuable information from datasets.[44] Hence,
large number of decision tree algorithms had
developed such as Quinlan’s ID3, C4.5, C5[44,45]
and Breiman et al., CART.[46] A large number
of tree classifiers had defined such as random
Indian Journal of Medical Sciences, Vol. 65, No. 6, June 2011
tree, rep tree, J 48, decision stump, MSP, LMT
and so on. But in this paper, decision stump
tree classifier is used to obtain desired result.
A  decision stump  (DS) tree defined as one
node decision tree that is based on stopping
the decision tree learner after the single most
informative feature.[47]
Logistic regression (LR)
Few of statistical algorithms are linear
discriminate analysis, least mean square
quadratic, kernel, logistic regression and k
nearest neighbors. But in this paper, Logistic
regression is used to obtain desired results.
Logistic regression is statistical classifiers
that are used for the analysis of data. It is
a type of linear regression that is used for
predicting binary or multi‑class‑dependent
variables.[48] Logistic regression (LR)[49] is a
classical classification method that has been
used widely in many applications including
document classification, [50,51] computer
vision, [52] natural language processing [53]
and bioinformatics. [54] LR can be defined
mathematically as Pr (G  =  k | X  =  x) is a
nonlinear function of x and range from 0 to 1
and sum up to 1.
Support vector machine (SVM)
Vapnik et al., [55] firstly used support vector
machines (SVM) to classification purpose.
But presently, SVMs have been used in a
wide range of problems including pattern
recognition, [56] bioinformatics [57] and text
categorization.[58] Hence, SVM classification
has done by realizing a linear or nonlinear
separation surface. But it can be found that
training of SVM requires solving quadratic
optimization problem.[59] A large number of
algorithms are proposed such as the sequential
 PREDICATION OF PARKINSON’S DISEASE
minimal optimization (SMO), nearest point
algorithm (NPA) etc., to solve this problem.[60,61]
Hence, in this paper, SMO classifier with
SVM is used to obtain desired result. Platt[60]
proposed this algorithm, called SMO for the
SVM classifier design.
Data source and performance matrices
Source
The dataset was created by Max little of the
University of Oxford, in collaboration with
the National Centre for Voice and Speech,
Denver, Colorado, who recorded the speech
signals. The original study published the feature
extraction methods for general voice disorders.
Dataset information
This dataset is composed of a range of
biomedical voice measurements from 31 people,
23 with Parkinson’s disease (PD). Each column
in the table is a particular voice measure,
and each row corresponds one of 195 voice
recording from these individuals (name column).
The main aim of the data is to discriminate
healthy people from those with PD, according
to “status” column which is set to 0 for healthy
and 1 for PD. The data are in ASCII CSV format.
The rows of the CSV file contain an instance
corresponding to one voice recording. There are
around six recordings per patient; the name of
the patient is identified in the first column.
Attribute information
Matrix column entries (attributes):
Name: ASCII subject name and recording
number
MDVP: Fo (Hz) ‑ Average vocal fundamental
frequency
MDVP: Fhi (Hz) ‑ Maximum vocal fundamental
frequency
Indian Journal of Medical Sciences, Vol. 65, No. 6, June 2011
235
MDVP: Flo (Hz) ‑ Minimum vocal fundamental
frequency
MDVP: Jitter(%), MDVP: Jitter (Abs), MDVP:
RAP, MDVP: PPQ, Jitter: DDP ‑ Several
Measures of variation in fundamental frequency
MDVP: Shimmer, MDVP: Shimmer (dB),
Shimmer: APQ3, Shimmer: APQ5, MDVP:
APQ, Shimmer: DDA ‑ Several measures of
variation in amplitude
NHR, HNR: Two measures of ratio of noise to
tonal components in the voice
Status: Health status of the subject (one) ‑
Parkinson′s, (zero) ‑ healthy
RPDE, D2: Two nonlinear dynamical
complexity measures
DFA: Signal fractal scaling exponent
Spread1, Spread2, PPE: Three nonlinear
measures of fundamental frequency variation
Performance matrices
In this paper, three performance matrices
are used to evaluate the performance of the
discussed classifiers/methods. Performance
of a diagnostic method is usually described in
terms of classification accuracy, sensitivity and
specificity.
• Accuracy: Accuracy defines relationship
between predicted value and actual value
i.e., how closes a predicted value to actual
value. Accuracy can be defined as
Accuracy → (TP + TN)/(TP + TN + FP + FN)
• Sensitivity: It can be defined as predicted
the value of output model with respect to
change in the input of model. It is used to
determine which attribute is more important
to obtained correct output value.[62] Now, it
can be defined as:
Sensitivity → TP/(TP + FN)
• Specificity: Specificity can be defined
236 INDIAN JOURNAL OF MEDICAL SCIENCES
as high degree of confidence. It can be
calculated as:
Specificity → TN/(TN + FP)
K‑FOLD CROSS VALIDATION METHOD
To obtain desired result, in this paper,
k‑fold cross validation method is used. This
method is also known as rotational method.
Basically, k‑fold cross method is derived
from cross‑validation method that is used to
measure and compare the learning algorithm.
In cross‑validation method data are divided in
two segments in which one segment is used
to learn/train a model and another used to
validate model. S. Larsen defined the idea of
cross‑validation in 1930.[63] Mosteller F. et al.,[64]
defined the statement for cross‑validation
that is similar to current k‑fold cross method.
Cross‑validation is appropriate for accuracy
predication and selection of model. [65‑67] In
k‑fold cross‑validation dataset is divided into
k disjoint folds of equal size d and learning
algorithm runs K times. Each iteration of
k‑fold cross method, different fold of data is
used for validation while k‑1 folds are used for
learning. Kohavi[68] provides comparative study
of several approaches to estimate accuracy
that includes cross‑validation (including
regular cross‑validation, leave‑one‑out
cross‑validation, and stratified cross‑validation)
and bootstrap (sample with replacement)
and recommended stratified 10‑fold
cross‑validation as the best model selection
method.
RESULT AND DISCUSSION
In this paper, the results are based on the
models that are evaluated by use of the
discussed classifiers i.e., decision stump (tree
Indian Journal of Medical Sciences, Vol. 65, No. 6, June 2011
classifier), logistic regression (statistical
classifier), and sequential minimization
optimization (support vector machine) with
the help of k‑fold cross‑validation method as
well as three performance matrices are used
to evaluate the performance of each model
for each fold. The results of these models are
represented in tabular as well as graphically.
Table 1 provides the tabular analysis of
discussed classifiers with help of k‑(10) fold
cross‑validation method. This table consists
of 10‑fold each model type and presented
the evaluated result in the form of confusion
matrix. A confusion matrix represents the
classification results. Confusion matrix defines
the relationship between value predicated by
the model and actual value.
Figure 1 define the confusion matrix
representation. The upper left cell of [Figure 1]
denotes the number of data classified as true
while they are true (i.e., true positives) and right
cell denotes the number of data classified as
true while they actually were false (i.e., false
positive). The lower left cell of [Figure 1] denotes
the number of data classified as false while
they actually were true (i.e., false negatives)
and lower right cell denotes the number of data
classified as false while they were actually false
(i.e., true false). After formation of confusion
matrix, the performance matrices i.e., accuracy,
sensitivity, and specificity are formed with
the help of discussed formula in performance
matrices section.
The analysis of Table 1 shows that model
formed with the help of decision stump
classifier provides the accuracy of 0.75 with
sensitivity of 0.91 and specificity of 0.13. The
LR model provides the accuracy of 0.64 with
 PREDICATION OF PARKINSON’S DISEASE 237

Table 1: Tabular analysis of k (10) folds cross‑validation method for all folds and models
Fold Decision stump (tree classifier) Logistic regression Sequential minimization
No. (function classifier) optimization(SVM classifier)
Confusion Acu. Sen. Spec. Confusion Acu. Sen. Spec. Confusion Acu. Sen. Spec.
matrix matrix matrix
1 142 5 0.76 0.97 0.15 100 47 0.65 0.68 0.54 142 5 0.75 0.97 0.10
41 7 22 26 43 5
2 142 5 0.75 0.90 0.13 104 43 0.69 0.71 0.65 144 3 0.75 0.98 0.06
33 15 17 31 45 3
3 142 5 0.75 0.93 0.12 96 91 0.53 0.51 0.60 142 5 0.76 0.97 0.15
38 10 19 29 41 7
4 142 5 0.75 0.90 0.13 96 51 0.65 0.65 0.65 144 3 0.78 0.98 0.17
33 15 17 31 40 8
5 142 5 0.75 0.93 0.12 99 48 0.66 0.67 0.60 142 5 0.75 0.97 0.10
38 10 19 29 43 5
6 142 5 0.75 0.90 0.13 96 52 0.64 0.65 0.63 142 5 0.76 0.97 0.13
33 15 18 30 42 6
7 142 5 0.75 0.90 0.13 93 54 0.63 0.63 0.63 144 3 0.78 0.98 0.17
33 15 18 30 40 8
8 142 5 0.75 0.90 0.13 97 50 0.65 0.66 0.60 144 3 0.76 0.98 0.10
33 15 19 29 43 5
9 142 5 0.75 0.90 0.13 97 50 0.66 0.66 0.65 144 3 0.78 0.98 0.17
33 15 17 31 40 8
10 142 5 0.75 0.90 0.13 97 50 0.66 0.66 0.65 143 4 0.76 0.97 0.13
33 15 17 31 42 6
Mean 0.75 0.91 0.13 0.64 0.64 0.62 0.76 0.9745 0.13
*Acu=→Accuracy, *Sen=→Sensitivity, *Spec=→Specificity

sensitivity of 0.64 and specificity of 0.62.
The model formed with help of sequential
minimization optimization i.e.,  SVM classifier
provides the accuracy of 0.76 with sensitivity of
0.97 and specificity of 0.13.
Figure 2 provides the accuracy parameter
comparison of model formed by decision stump
(DS), Logistic regression (LR) and sequential
minimization optimization (SMO) classifiers.
From [Figure 2], it is clear that LR model has
poor performance among these models i.e.,
0.64 (accuracy result), while the performance
of SMO model is best among these models i.e.,
0.76. Because LR model identified 64 people
out of 100 people with PD affected while in
case of SMO model, it indicates that 75 people
out of 100 with PD affected.
Figure 3 shows the comparison of sensitivity
parameter which provides the similar result
to accuracy parameter. Hence the analysis
of [Figure 3] shows that the LR model
provides worst result i.e., 0.64 while SMO
model provides best result among discussed
classifiers i.e., 0.97.
But the analysis of specificity parameter
[Figure 4] provides just opposite result to
accuracy [Figure 2] and sensitivity [Figure 3]
parameter. Because in case of specificity
parameter [Figure 4], DS and SMO models
have similar performance i.e., 0.13 while the
performance of LR model is 0.62.
But, the model formed with a high sensitivity,
but low specificity i.e., SMO model (sensitivity

Indian Journal of Medical Sciences, Vol. 65, No. 6, June 2011

238 INDIAN JOURNAL OF MEDICAL SCIENCES


  


$FFXUDF\
 

'6 /5 602

Figure 2: Accuracy comparison of decision stump, logistic

Figure 1: Confusion matrix representation regression and sequential minimization optimization
classifiers


  
   
 

 6HQVLWLYLW\ 
6SHFLILFLW\
 
  

  
'6 /5 602
'6 /5 602
Figure 3: Sensitivity comparison of decision stump,
Figure 4: Specificity comparison of decision stump,
logistic regression and sequential minimization
logistic regression and sequential minimization
optimization classifiers
optimization classifiers

0.97 and specificity 0.13) and DS person is represented with 0. To obtain

model  (sensitivity 0.91 and specificity 0.13)
results in many patients who are disease free,
may have possibility the disease and subject to
further investigation.
CONCLUSION
In this paper, we try to develop some
predication model for Parkinson’s disease
identification. For this purpose, three data
mining methods i.e., decision stump (tree
classifiers), logistic regression (statistical
classifiers) and sequential minimization
optimization (support vector machine) are
used. Dataset that is used in this paper has
taken from UCI repository. This dataset is
composed of a range of biomedical voice
measurements from 31 people, 23 with
Parkinson’s disease (PD). Each column in the
table is a particular voice measure, and each
row corresponds one of 195 voice recording
from these individuals. PD affected person
is represented with a value of 1 and healthy
the desired result, 10 cross fold method is
used with discussed classifiers as well as
three parameters are used to analyze the
performance of discussed classifiers. These
three parameters include accuracy, sensitivity,
and specificity. The aggregate results indicate
that in terms of accuracy and sensitivity,
support vector machine model performed very
well i.e., accuracy 0.76 and sensitivity 0.97
and statistical model i.e., logistic regression
performed worst i.e., accuracy 0.64 and
sensitivity 0.64. But the specificity parameter
indicates that statistical model of 0.62 doing
well as compare another two models (tree
model of 0.13 and support vector machine
model of 0.13). But when the result of these
models is compared with real‑world results
that are defining in our data source i.e.,  23
people out of 31(almost 75 out of 100) are PD
affected. Hence, the number of people with PD
in case of tree and SVM classifier is almost
23, while in case of statistical classifier  (LR)
is 20. Hence, we can easily say that tree

Indian Journal of Medical Sciences, Vol. 65, No. 6, June 2011

 PREDICATION OF PARKINSON’S DISEASE
and SVM classifiers accurately identified the
number of people with PD. But analyzing the
performance of the sensitivity and specificity
parameter, there is large difference in case
of tree and SVM method i.e., 0.78 and 0.84
but in case of statistical method, performance
is almost same i.e., 0.64 and 0.62. Hence,
the model with a high sensitivity but low
specificity results in many patients who are
disease free, may have possibility the disease
and subject to further investigation. Hence, in
case of tree and SVM method sensitivity (0.91
and 0.97) is high but specificity (0.13 and
0.13) is low further investigation of people is
required but in case of LR model sensitivity
and specificity is almost same i.e., 0.64
and 0.62 and there is no need of further
investigation. Hence, we finally conclude that
LR model indentified people with PD more
correctly than tree and SVM classifiers on the
behalf of discussed performance matrices.
So in the study of Parkinson’s disease, only
voice measurements of people have been
considered to identify the person with PD.
There are lot of symptoms that lead the
Parkinson’s disease such as age factor,
environmental factor, trembling in the hands,
arms, legs, impaired speech production and
speech articulation difficulties. But in this
paper, speech articulation difficulty of PD
affected people is considered for formation of
the model and analyzes the discussed model
on this symptom of Parkinson’s disease.
REFERENCES
1. Elbaz A, Bower JH, Peterson BJ, Maraganore DM,
McDonnell SK, Ahlskog JE, et al. Survival
Study of Parkinson Disease in Olmsted County,
Minnesota. Arch Neurol 2003;60:91‑6.
Indian Journal of Medical Sciences, Vol. 65, No. 6, June 2011
239
2. Parkinson J. An essay on the shaking palsy.1817.
J Neuropsychiatry Clin Neurosci 2002; 14:223‑36.
3. Tanner CM, Ross GW, Jewell SA, Hauser RA,
Jankovic J, Factor SA. Occupation and risk of
Parkinsonism: A multicenter case‑ control study.
Arch Neurol 2009;66:1106‑13.
4. Marras C, Tanner C. Epidemiology of Parkinson’s
Disease. Movement Disorders. In: Watts RL,
Koller WC, editors Neurologic Principles and
Practice, 2nd ed. New York: The McGraw‑Hill
Companies; 2004. p. 177.
5. US Census Bureau. US interim projections by
age, sex, race, and Hispanic origin: 2000‑2050.
Available from: http://www.census.gov/population/
ww/projections/usinterimproj. [Last Accessed on
2012 Apr. 7].
6. Available from: http://www.rightdiagnosis.com/p/
parkinsons_disease/stats‑country.htm. [Last
Accessed on 2012 Apr. 7].
7. Dauer W, Przedborski S. Parkinson’s disease:
Mechanisms and Models. Neuron 2003;39:
889‑909.
8. Alonso JB, de Leon J, Alonso I, Ferrer MA.
Automatic detection of pathologies in the voice
by HOS based parameters. EURASIP J Appl Sig
process 2001;14:275‑84.
9. Cnockaert L, Schoentgen J, Auzou P, Ozsancak
C, Defebvre L, Grenez F. Low‑ frequency vocal
modulations in vowels produced by Parkinsonian
subjects. Speec Commun 2008;50:288‑300.
10. Revett K, Gorunescu F, Mohamed Salem AB.
Feature Selection in Parkinson’s disease: A rough
Sets approach. Proceedings of the International
Multi conference on Computer Science and
Information Technology; Oct. 12‑14 Margowo,
Polond 2009;4: p. 425‑8.
11. Anthony E, Lees LA. Management of Parkinson’s
disease: An evidence‑based review. Mov Disord
2002; 17(Suppl 4):S1‑166.
12. B l o e m   B R , B e c k l e y   D J , v a n D i j k   J G ,
Zwinderman AH, Remler MP, Roos RA. Influence
of dopaminergic medication on automatic postural
responses and balance impairment in Parkinson’s
240 INDIAN JOURNAL OF MEDICAL SCIENCES
disease. Mov Disord 1996;11:509‑21.
13. Hunt E, Martin J, Stone P. Experiments in
Induction. New York: Academic Press, 1966.
14. Nilsson N. Learning Machines: Foundations of
Trainable Pattern Classifying Systems, New York:
McGraw‑Hill, 1965.
15. Rosenblatt F. Principles of Neurodynamics.
Washington, DC: Spartan Books; 1962.
16. Lavrac N, Keravnou E, Zupan B, editors (1997).
Intelligent data analysis in medicine and
pharmacology. Boston: Kluwer.
17. Lucas PJF, Abu‑Hanna A. Prognostic methods in
medicine. Artif Intell Med 1999;15:105‑19.
18. Maclin PS, Dempsey J, Brooks J, Rand J. Using
neural networks to diagnose cancer. J Med Syst
1991;15:11‑9.
19. Cicchetti DV. Neural networks and diagnosis in
the clinical laboratory: State of the art. Clin Chem
1999;38:9‑10.
20. Pendharkar PC, Rodger JA, Yaverbaum GJ,
Herman N, Benner M. Association, statistical,
mathematical and neural approaches for mining
breast cancer patterns. Expert Syst Application
1999;17:223‑32.
21. Delen D, Walker G, Kadam A. Predicting breast
cancer survivability: A comparison of three data
mining methods. Artif Intell Med 2005;34:113‑27.
22. C o o p e r   G F , A l i f e r i s   C F , A m b r o s i n o   R ,
Aronis J, Buchanan BG, Caruana R, et al. An
evaluation of machine‑learning methods for
predicting pneumonia mortality. Artif Intell Med
1997;9:107‑38.
23. Majumder SK, Ghosh N, Gupta PK. Support vector
machine for optical diagnosis of cancer. J Biomed
Opt 2005;10:024034.
24. Gokturk SB, Tomasi C, Acar B, Beaulieu CF,
Paik DS, Jeffrey RB. A statistical 3‑D pattern
processing method for computer‑aided detection
of polyps in CT colonography. IEEE Trans Med
Imaging 2001;20:1251‑60.
25. El‑Naqa I, Yang Y, Wernick MN, Galatsanos NP,
Nishikawa RM. A support vector machine
approach for detection of micro calcifications.
Indian Journal of Medical Sciences, Vol. 65, No. 6, June 2011
IEEE Trans Med Imaging 2002;21:1552‑63.
26. Scholkopf B, Tsuda K, Vert JP. Kernel Methods
in Computational Biology. Cambridge, MA:
MIT.2004, In Press.
27. Wei L, Yang Y, Nishikawa RM, Jiang Y. A study
on several machine learning methods for
classification of malignant and benign clustered
micro calcifications. IEEE Trans Med Imaging
2005;24:371‑80.
28. Andreassen S, Woldbye M, Falck B, Andersen SK.
MUNIN‑A causal probabilistic network for
interpretation of electromyrographic findings.
Proceedings International Joint Conference
Artificial Intelligence. 10th ed. J McDermott, Los
Altos, CA: Morgan Kaufmann; 1987; p. 366‑72.
29. Diez FJ, Mira J, Iturralde E, Zubillaga S. DIVAVAL,
a Bayesian expert system for echocardiography
Artif Intell Med 1997;10:59‑73.
30. Friedman N. Inferring cellular networks using
probabilistic graphical models. Science
2004;303:799‑805.
31. Nikovski D. Constructing bayesian networks for
medical diagnosis from incomplete and partially
correct statistics. IEEE Trans Knowledge Data
Eng 2007;12:509‑16.
32. Petricoin EF, Liotta LA. SELDI‑TOF‑based serum
proteomic pattern diagnostics for early detection
of cancer. Curr Opin Biotechnol 2004;15:24‑30.
33. Bocchi L, Coppini G, Nori J, Valli G. Detection
of single and clustered microcalcifications in
mammograms using fractals models and neural
networks. Med Eng Phys 2004;26:303‑12.
34. Dettling M. Bag Boosting for tumor classification
with gene expression data. Bioinformatics
2004;20:3583‑93.
35. Wang JX, Zhang B, Yu JK, Liu J, Yang MQ, Zheng S.
Application of serum protein fingerprinting coupled
with artificial neural network model in diagnosis
of hepatocellular carcinoma. Chin Med J (Engl)
2005;118:1278‑84.
36. Zhou X, Liu KY, Wong ST. Cancer classification
and prediction using logistic regression with
Bayesian gene selection. J Biomed Inform
 PREDICATION OF PARKINSON’S DISEASE
2004;37:249‑59.
37. O’Reilly F, Finnan F, Allwright S, Smith GD,
Ben‑Shlomo Y. The effects of caring for a
spouse with Parkinson’s disease on social,
psychological and physical well‑being. Br J Gen
Pract 1996;46:507‑12.
38. K e u s   S H , B l o e m   B R , H e n d r i k s   E J ,
Bredero‑Cohen AB, Munneke M. Practice
Recommendations Development Group.
Evidence‑Based Analysis of Physical Therapy
in Parkinson’s Disease with Recommendations
for Practice and Research. Mov Disord
2007;22:451‑60.
39. Khan F, Amatya B. Rehabilitation for Parkinson’s
disease: Analysis of the Australian rehabilitation
outcomes dataset. J Clin Med Res 2010;3:1‑8.
40. De Boer AG, Wijker W, Speelman JD, De Haes JC.
Quality of life in patients with Parkinson’s disease:
Development of a questionnaire. J Neurol
Neurosurg Psychiatry 1996;61:70‑4.
41. De Boer AG, Wijker W, Bartelsman JF,
de Haes HC. Inflammatory bowel disease
questionnaire: Crosscultural adaptation and
further validation. Euro J Gastroentero Hepatol
1995;7:1043‑550.
42. Limousin P, Karck P, Pollak P, Benazzouz A,
Ardouin C, Hoffmann D, et al. Electrical
stimulation of the subthalamic nucleus in
advanced parkinson’s disease. N Eng J Med
1998;339:1105‑11.
43. Yunfeng Wu, Krishna S. Statistical analysis of gait
rhythm in patients with Parkinson’s Disease. IEEE
Tran Neural Sys Rehabil Engin 2007;18:150‑18.
44. Quinlan JR. C4.5: Programs for Machine Learning.
San Francisco, USA: Morgan Kaufmann 1995.
45. Quinlan J. Induction of decision trees. Mach Learn
1986;1:81‑106.
46. Breiman L, Friedman JH, Olshen RA, Stone CJ.
Classification and regression trees. Monterey,
CA: Wadsworth and Brooks/Cole Advanced
Books and Software; 1984.
47. Holte R. Very simple classification rules perform
well on most commonly used dataset. Mac learn
Indian Journal of Medical Sciences, Vol. 65, No. 6, June 2011
241
1993;11:63‑91.
48. Hastie T, Tibshirani R, Friedman J. The
elements of statistical learning. New York, NY:
Springer‑Verlag; 2001.
49. Minka TP. A comparison of numerical optimizers
for logistic regression. Technical report,
Technical Report, Department of Statistics,
Carnegie Mellon University, Pittsburgh, PA, USA,
2003(revised2007).
50. Friedman J, Hastie T, Tibshirani R. Additive
logistic regression: A statistical view of boosting.
Ann Statis 2000;28:337‑407.
51. Maghbouleh A. A logistic regression model
for detecting prominences. In The Fourth
International Conference on Spoken Language,
ACM, Oct. 3 rd - 6 th, Philadelphia, PA, USA,
1996;4:2443‑5.
52. Liao JG, Chin K. Logistic regression for disease
classification using microarray data. Bioinformatics
2007;23:1945‑51.
53. Zhu J, Hastie T. Kernel logistic regression and
the import vector machine. J Comput Graph Stati
2001; 14:1081‑88.
54. Zhu J, Hastie T. Classification of gene microarrays
by penalized logistic regression. Biostatistics
2004;5:427‑43.
55. Vapnik V N. The nature of statistical learning
theory. New York: Springer‑Verlag; 1995.
56. Pontil M, Verri A. Support vector machines for
3D object recognition. IEEE Trans Patt Anal Mac
Intell 1998;20:637‑46.
57. Yu GX, Ostrouchov G, Geist A, Samatova NF.
An SVM‑based algorithm for identification of
photosynthesis‑specific genome features.
In proceeding of 2 nd IEEE computer society
bioinformatics conference, IEEE, 11‑ 14 August
Stanford CA, USA 2003; 2:235‑43.
58. Joachim’s T. Text categorization with support
vector machines. In Proceedings of 10th European
conference on machine learning (ECML),
Springer, 21st‑23rd April, Chemintz, Germany
1998. p. 137‑ 42.
242 INDIAN JOURNAL OF MEDICAL SCIENCES
59. Vishwanathan SVM, Murty, Narasimha M. SSVM:
A Simple SVM Algorithm. In: Proceedings of
the International Joint Conference on Neural
Network, IEEE, 12-17 May Honolulu, Hawaii
2002:3:2393‑8.
60. Platt JC. Advances in Kernel Methods: Support
Vector Machines. In: Schölkopf B, Burges C,
Smola A, editors. Fast training of support vector
machines using sequential minimal optimization.
Cambridge, MA: MIT; 1998, In Press.
61. Keerthi SS, Shevade SK, Bhattacharyya C,
Murthy KK. A fast iterative nearest point algorithm
for support vector machine classifier design. IEEE
Trans Neural Netw 2000;11:12436.
62. Saltelle A, Chan K, Scott EM. Sensitivity Analysis:
Gauging the Worth of Scientific Models. John
Wiley and Sons; Edition 1: 2000; p. 492.
63. Larson S. The shrinkage of the coefficient
of multiple correlations. J Educat Psychol
1931;22:45‑55.
64. Mosteller F, Turkey JW. Data analysis, including
statistics. In: G. Lindzey, E. Aronson, editors.
Handbook of Social Psychology. 2 nd  ed.
Indian Journal of Medical Sciences, Vol. 65, No. 6, June 2011
View publication stats
Addison‑Wesley, Reading, MA, 1968. p. 80‑230.
65. Breiman J, Friedman R, Stone OC. Classification
and Regression Trees. Belmont CA: Wadsworth,
1984.
66. Breiman L, Spector P. Sub model selection and
evaluation in regression. The X‑random case.
Internat Statist Rev 1992;60:291‑319.
67. Schaffer C. Selecting a classification method by
cross‑validation. Mac Learn 1993;13:135‑43.
68. Kohavi R. A study of cross‑validation and bootstrap
for accuracy estimation and model selection. In
Proceedings of 14th International Joint Conference
on AI, Morgan Kaufmann 20‑25 August Canada
1995;2:1137‑45. Available from: http://citeseer.
ist.psu.edu/kohavi95study. [Last accessed on
2012 Apr 4].
How to cite this article: Yadav G, Kumar Y, Sahoo G. Predication
of Parkinson’s disease using data mining methods: A comparative
analysis of tree, statistical, and support vector machine classifiers.
Indian J Med Sci 2011;65:231-42.
Source of Support: There is no financial source of support to
carry out the work.. Conflict of Interest: In this paper, there is no
conflict of interest. I hereby take full responsibility for the veracity
of the information provided on behalf of all coauthors.