Parkinson Disease Prediction Using
Machine Learning Algorithm
Richa Mathur, Vibhakar Pathak and Devesh Bandil
Abstract Parkinson disease, the second most common neurological disorder that
causes significant disability, reduces the quality of life and has no cure.
Approximately, 90% affected people with Parkinson have speech disorders. The
medical dataset contains heterogeneous data in the form of text, numbers, and
images that can be mined. Big Data has the potential to give valuable information
after processing that can be discovered through deep analysis and efficient pro-
cessing of data by decision-makers. Data mining is the process of selecting,
extracting, and modeling the unknown hidden patterns from large datasets. Machine
learning algorithm (MLA) can be used for early detection of disease to increase the
chances of elderly people’s lifespan and improved lifestyle with Parkinson. In this
paper, we use various MLAs that can help in improving the performance of datasets
and play a vital role in making the early prediction of disease at right time. After
comparison of these algorithms, we choose the most effective one in terms of
accuracy. From our experimental results, it is analyzed that the accuracy obtained
from the combined effect of KNN algorithm with ANN is better as compared to
other algorithms.
Keywords Parkinson disease Predictive analytics Voice datasets
SVM KNN ANN
R. Mathur (&)
Department of CS & Applications, SGVU, Jaipur, Rajasthan, India
e-mail: [email protected]
V. Pathak
CS/IT, Arya College of Engineering and IT, Jaipur, Rajasthan, India
e-mail: [email protected]
D. Bandil
Computer Applications, Suresh Gyan Vihar University, Jaipur, Rajasthan, India
e-mail: [email protected]
© Springer Nature Singapore Pte Ltd. 2019 357
V. S. Rathore et al. (eds.), Emerging Trends in Expert Applications and Security,
Advances in Intelligent Systems and Computing 841,
https://doi.org/10.1007/978-981-13-2285-3_42
358 R. Mathur et al.
1 Introduction
Parkinson disease (PD), a neurodegenerative disorder of CNS system, also pro-
duces movement disorder. In the primary stage when nerve cells or neurons of the
brain become impaired, people may begin to notice symptoms like tremor, stiffness
in limb or trunk of the body, movement issues, or impaired balance. With the
progression of the disease, people may have difficulty in walking, talking, or
completing other simple tasks. PD has no cure but several treatments are known to
provide relief from the symptoms. PD affects brain cells that produce Dopamine, a
neurotransmitter, which is responsible for coordinate and control muscle
activity [1].
Typically, PD occurs in people over the age of 60, among which 1% of people
are affected. It is called as young onset PD when it is seen in the people before age
50 [2]. According to estimates, PD affected 6.2 million people and about 117400
deaths globally in 2015 [3].
Nowadays, data is becoming more valuable but how to handle data and finding
hidden facts from it is more important. The term “Big Data” describes a large
amount of datasets that are so complex that it is not possible to process them via
conventional methods and technologies. To extract valuable insights from such
varied and rapid growing datasets, various tools and techniques of Big Data ana-
lytics can be used that may lead to better decision-making and strategic planning.
The rising population generates a large amount of data related to patients clinical
and laboratory tests. With this dataset, doctors can detect and diagnose the disease
at their early stages. Early prediction of motor symptoms of PD can get a proper
treatment at right time to a patient.
2 Related Work
Shamli and Sathiyabhama [4] proposed multi-classifier system, i.e., based on Big
Data analytics to improve predictive performance and efficient time to answer
cost-effective actions. The author introduced Big Data with its characteristics and
Big Data analytics with their types as Descriptive, Predictive and Prescriptive in
healthcare industries. Dopamine, a neurotransmitter, generated by brain cells, is
responsible to send signals to other brain cells to control muscle activity. The
degeneration of dopamine-producing brain cells causes PD. For analysis purposes,
voice dataset of PD is collected from UCI machine learning library. By imple-
menting multiple predictive models to disease datasets, multiple accuracies and
results of different classifiers are acquired. C4.5, SVM, and ANN give better results
than other machine learning algorithms. After comparing the results of these
classifiers, best results are chosen for the final decision. This approach helps
organizations to analyze their large datasets quickly and efficiently with maximum
accuracy.
Parkinson Disease Prediction Using Machine Learning Algorithm 359
Azad et al. [5] explored a predictive model for PD that is based on decision tree
algorithm. They introduced PD, a second most common neurodegenerative disease
with its symptoms, possible complications, and risk factors associated with it.
Various applications of data mining are used for classification purposes that are
decision tree, attribute selection measures, ID3 and decision stumps. Their dataset
(have 197 instances) is taken from UCI repository and built up from the data of 31
people. For performance analysis, two parameters accuracy and classification error
are used. For validation, 10-fold cross-validation technique is used that gives the
unbiased outcome. They found that decision tree algorithm performs best and gives
the best accuracy and less classification error than other algorithms in their
experimental results.
Sriram et al. [6] proposed a method for diagnosis of PD using its voice dataset.
This voice dataset is built up from the voice of 31 people among which 23 people
are affected by PD. This dataset contains 5875 instances and 26 attributes. In their
experiment for statistical analysis, classification, evaluation, visualization, and
unsupervised methods Weka V3.4.10 and Orange V2.0b software are used. They
achieved the best accuracy 90.2% from Random Forest algorithm.
3 Problem Definition
The huge amount of data known as Big Data is generated everywhere, and this data
can be used to perform analysis and make future predictions. According to some
research, most of the healthcare data is in the unstructured form that can be stored in
the centralized repository to make useful interpretation out of it. To improve the
quality of patient care at low cost, this unstructured data can be analyzed further by
merging it with structured datasets. The problem is to classify and discover data
pattern to predict future disease, so that doctors can detect and diagnose the disease
at an early stage.
4 System Architecture
The dataset used in this paper is taken from UCI library [7]. Analysis of these data
will provide early diagnosis and detection of disease at reducing cost [4]. The
gathered information is in unstructured format, i.e., it is not in a particular kind of
format. After that, we convert this mixed type of disease data into the structured
form which will ease the process.
For that, a layered Big Data framework is used which has mainly following three
components:
Hadoop: A very popular, distributed processing, and storage framework that can
handle large and complex unstructured data. Therefore, Hadoop is the best option
360 R. Mathur et al.
for analyzing unstructured disease dataset. Hadoop has two main components:
MapReduce and HDFS for processing and storing a large amount of datasets.
Hadoop uses HDFS to store very large data files (in GBs to TBs) that cannot be
stored on a single machine. MapReduce is a software programming paradigm used
to process a large amount of datasets by distributing the work to various inde-
pendent nodes.
Predictive Analytics: Is a probabilistic platform for predicting the future. It uses a
variety of statistical modeling, data mining techniques, and machine learning
techniques. It provides actionable insights based on the data so that organizations
can identify a pattern from the data and apply statistical modeling technique and
algorithms to find relationships between various datasets [4].
Prediction Models: To classify data various machine learning (ML) algorithms are
used. ML (i.e., a branch of Artificial Intelligence (AI) concerned with the study of
classification and pattern analysis) algorithm allows us to automatically recognize
complex patterns and make intelligent decisions based on data. Some ML algo-
rithms that we used in this paper are as follows:
Support Vector Machine (SVM) yields more accurate results when it is used for
classifying text. It is successfully employed in text classification and various other
sequence processing applications as it is a type of linear classifier.
Artificial Neural Network (ANN) is a type of supervised learning models and it
is derived from the functionality of human brains, i.e., highly sophisticated ana-
lytical technique, capable of modeling extremely complex nonlinear functions [8].
We used a popular ANN algorithm called multilayer perceptron (MLP), i.e., a type
of supervised learning model used for prediction and classification problems [9].
5 Proposed Model
We use Waikato Environment for Knowledge Analysis (WEKA) to implement data
mining algorithms for preprocessing, classification, clustering, and analysis of
results. This environment includes java libraries that implement algorithms and
provide the best environment to researchers for classifying datasets.
5.1 Data Collection
The dataset used in this paper is taken from UCI machine learning library [7]. The
dataset consists of 195 instances and 24 attributes. This feature set consists of name,
Fo(Hz), Fhi(Hz), Flo(Hz), Jitter(%), Jitter(Abs), RAP, PPQ, Jitter:DDP, Shimmer,
Shimmer(dB), Shimmer:APQ3, Shimmer:APQ5, APQ, Shimmer:DDA, NHR,
HNR, status, RPDE, DFA, spread1, spread2, D2, and PPE. We store these datasets
in .CSV format and then convert it into. ARFF format for further analysis. The
Parkinson Disease Prediction Using Machine Learning Algorithm 361
dataset is divided into two classes according to its “status” column which is set to 0
for healthy subjects and 1 for those PD [10].
5.2 Data Preprocessing
The poor data quality in the medical dataset is one of the big challenges that are
faced by the knowledge discovery process. This process decreases the number of
attributes into a better subset which can increase accuracy, and also it brings a
reduction in training time. It is done using Filters and Wrappers. WEKA provides
“AttributeSelection” filter to choose an attribute evaluation method. We use
“cfsSubsetEval” attribute evaluator and “BestFirstSearch” method which considers
the individual predictive ability of each feature to evaluate the worth of an attribute.
From that, a new feature data subset is prepared which contains 11 features.
5.3 Data Mining
In this proposed framework, we used different classification techniques for ana-
lyzing PD patient’s record. To evaluate performance, we apply 10-fold
cross-validation technique which splits the original set into training sample to
train the model and a test set to evaluate results.
An approach known as “Information Retrieval Metrics” can be used to evaluate
experimental results in terms of precision, recall, f-measure, and accuracy with the
use of following formulas [11]:
Precision : TP=ðTP þ FPÞ; Recall : TP=ðTP þ FPÞ
F measure : 2 Precision Recall=ðPrecision þ RecallÞ
Accuracy : TP þ TN=ðTP þ TN þ FP þ FNÞ
Here, TP = True Positive; TN = True Negative; FP = False Positive;
FN = False Negative [12].
6 Experimental Results
Various techniques are used in the analysis and prediction of PD. Methods that are
based on analytics can give an appropriate prediction for a particular disease by
grouping people with similar symptoms. In our experiment, obtained accuracies
using SMO, KNN [13], Random Forest, AdaBoost.M1 [14], Bagging, MLP, and
DT algorithms are 86.67%, 90.76%, 89.23%, 88.20%, 89.23%, and 89.74%,
362 R. Mathur et al.
Table 1 Performance measured by classifiers
KNN + AdaBoosta.M1 KNN + Bagging KNN + MLP
Accuracy 91.28% 90.76 91.28
Classification error 8.717 9.23 8.71
Time taken to build model 0.01 0.02 0.43
Precision 0.873 0.866 0.873
Recall 0.907 0.904 0.907
F-Measure 0.888 0.882 0.888
respectively. In our experiment, we used several ensemble methods that are capable
to combine classifiers with their predictions and base estimators. For using more
than one classification model, we have to “meta” option under classification tab,
and then select “vote” classifier. After that, select the classifier properties and in
classifiers tab, we can add multiple classifiers as per our need. KNN provides better
accuracy and less execution time than SMO and random forest. So we combined
ANN algorithms with KNN algorithm. Table 1 shows performance measure that is
reported by our experimental result with disease dataset, after conducting 10-fold
cross-validation technique.
Figure 1 shows overall precision, recall, and F-measure rate of combined
approach. The accuracy acquired by AdaBoosta.M1 and MLP with KNN are same,
i.e., 91.28% which is better when we use these separate algorithms. Our experi-
mental result shows that same preprocessing methods on a different dataset affect
similarly the classifiers performance. After analyzing results, it is observed that
when we combine two classifiers, accuracy is increased and time taken to build
model is reduced. The accuracy acquired by AdaBoosta.M1 and MLP with KNN
are same, i.e., 91.28% which is better when we use these separate algorithms. And
this accuracy becomes 100% when we use training dataset with selected attributes.
Time taken to build a model with AdaBoosta is 0.01 s, whereas time taken to
build model with MLP is 0.43 s. So that AdaBoosta1 with KNN gives best accuracy
with time and less classification error.
Fig. 1 Comparison of 100
accuracy achieved by ANN 80
algorithms with KNN Accuracy
60
40 Precision
20 Recall
0
Parkinson Disease Prediction Using Machine Learning Algorithm 363
7 Conclusion and Future Scope
Big Data analytics plays a huge role in the healthcare industry, as these data are
scattered everywhere, big, and complex in nature. In this paper, we discuss early
stage prediction of Parkinson disease for that we presented a methodology of data
mining using Weka tool for classifying disease dataset. We use various MLAs for
classifying our experimental data that indicate the combined effect of ANN algo-
rithms with KNN which is better as compared when we use other algorithms. The
system detects the maximum accuracy of the multi-classifier, and their result pre-
dicts the disease at its early stage. We discuss the comparative analysis and cal-
culate the overall performance measures in terms of precision, recall, and
f-measure. In future, effective optimization techniques can be used to achieve
better accuracy and cost-effective interventions for Parkinson disease. Also, limited
data is available that describes the real potential of early PD treatment which
requires more research to explore the real impact of early treatment.
References
1. https://en.wikipedia.org/wiki/Parkinson%27s_disease
2. PD: hope through research. https://www.ninds.nih.gov/Disorders/Patient-Caregiver-Education/
Hope-Through-Research/Parkinsons-Disease-Hope-Through-Research
3. Saloni et al (2015) Detection of Parkinson disease using clinical voice data mining
4. Shamli N et al (2016) Parkinson’s Brain disease prediction using big data analytics
5. Azad C et al (2014) Design and analysis of data mining based prediction model for
Parkinson’s disease
6. Sriram TVS et al (2013) Intelligent Parkinson disease prediction using machine learning
algorithms
7. Shaikh, TA (2014) A prototype of Parkinson’s and Primary tumor diseases prediction using
data mining techniques
8. Kirubha V et al (2016) Survey on data mining algorithms in disease prediction
9. Salekin A Detection of chronic Kidney disease and selecting important predictive attributes
10. Gaur V et al (2013) A multi-objective optimization of cloud based SLA-Violation prediction
and adaptation
11. PD dataset from UCI repository. https://archive.ics.uci.edu/ml/machine-learning-databases/
parkinsons/
12. Boukenze B et al (2016) Performance on data mining techniques to predict in healthcare
industry: Chronic Kidney failure disease
13. Rana M et al (2015) Breast Cancer diagnosis and recurrence prediction using machine
learning techniques
14. Freund Y et al (1996) Experiments with a new boosting algorithm
