A clustering algorithm application in Parkinson disease based on k-means method

Full text

International Journal of Mathematics and
Computer Science, 15(2020), no. 4, 1005–1014
b b
M
CS
A Clustering Algorithm Application in
Parkinson Disease based on k−means Method
Israa Ali Alshabeeb, Nidaa Ghalib Ali,
Saba Abdulameer Naser, Wafaa M. R. Shakir
Technical Computer System Department
Babylon Technical Institute
Al-Furat Al-Awsat Technical University
Babil, Iraq
Inb.esr@atu.edu.iq, Inb.nedaa10@atu.edu.iq,
sabaabdulameernaser@gmail.com, inb.wfa@atu.edu.iq
(Received July 21, 2020, Accepted August 17, 2020)
Abstract
Data mining methods are used to predict and compare ages of peo-
ple with Parkinson’s disease and so are considered a critical part in
the medical community. The amount of data stored in the medical
section database is increasing rapidly. There is a system used to ar-
range the results of ages of Parkinson disease patients by years. In this
paper, a data mining technique called the k−means clustering algo-
rithm for analyzing data is implemented. This technique is applied on
a hospital database and analyzes the performance successfully and so
is useful in giving accurate results and making an effective decision by
the Ministry of Health in Iraq and related parties to find appropriate
solutions.
1 Introduction
The ability to monitor the performance of a disease in a huge number of pa-
tients is becoming of increasing importance. It is crucial to know how many
Key words and phrases: Data Mining, Clustering, Parkinson disease,
k−means clustering.
AMS (MOS) Subject Classifications: 68Q25, 68W40, 62H30
ISSN 1814-0432, 2020, http://ijmcs.future-in-tech.net

1006 I. A. Alshabeeb, N. G. Ali,S. A. Naser, W. M. Shakir
people are affected by diseases in relevance to different ages. A second popu-
lar degenerative disorder disease after Alzheimer’s is Parkinson disease (PD).
The most common symptoms of PD are shaking and slowness of movement.
The main cause of this disease is unknown [1] [2]. In data mining, many
algorithms have been used with diseases. Decision tree, factor analysis, neu-
ral net and logistic regression were implemented to examine the biomedical
voice measurements to find out which measurement is more suitable to figure
out the early symptoms of PD and follow early treatment [3]. Using data
mining techniques such as Decision Tree Algorithm, Naive Bayes and Neural
Network doctors can detect the risk rate of a heart disease [4]. Deepika and
Kalaiselvi [5] gave a review of using data mining methods and comparing the
results for diagnosis and prognosis of breast cancer disease, heart disease and
thyroid. Using technical regression, the decision tree models and technical
regression to predict diabetes using specific risk factors were discussed in [6].
While Bayesian analysis was used to estimate how PD depends on hereditary
as a risk factor key and how it affected PD patients [7]. In this work, a new
approach is used to study yearly comparisons of PD patients to see if the
disease is growing or not. The algorithm that is proposed in this paper gives
good accuracy.
Physicians and The Ministry of Health in Iraq try to know how a disease has
been progressing to find solutions to treat people at an early stage. There are
many clustering methods in data mining that are available to predict how
many patients are affected every year. The k−means clustering method is
used in this paper to know how PD affects people.
2 Methodology
2.1 Clustering
Clustering is a data mining technique for gathering similar objects in fea-
tures or properties in one group which is called a cluster. When the distance
between two objects is less than any other distance of other objects will be
in the same subset and this subset should contain at least one object [8].
This technique simply puts similar data into a groups and dissimilar ob-
jects into a separate group [9]. Clustering is widely used in many different
fields and applications such as pattern recognition, image processing, secu-
rity, machine-learning situations, data analysis, business, web search and
biology [10]. Clustering methods can be classified in many types. Those
are represented by partitioning methods, density-based methods, hierarchi-

A Clustering Algorithm Application in Parkinson ... 1007
Figure 1: Partitioning clustering
cal methods, grid-based methods and model-based methods [11]. The type
of data and the purpose of application are the main points to choose the best
clustering method. Many researches have used clustering algorithms, like Hi-
erarchical clustering [12], k−means clustering [13] [14] in applications. Par-
titioning methods work based on iterative reallocating data objects among
subsets and determine the initial number of subsets. Partitioning algorithms
are represented by the k−medoids and k−means. k−means algorithm is
proposed in this paper. Partitioning clustering is shown in figure 1
2.2 Data mining process
In this work, data was gathered from Neuroscience hospital in Baghdad, Iraq
for four years (2016-2019) and a classification method to analyze them.
2.2.1 k−means Method
k−means is one of the clustering algorithms. It is used to cluster the data
into groups based on a centroid point. This algorithm is used widely in
pattern recognition applications [15]. It is useful with both a big or small
dataset and gives a good result. k−means clustering is considered as an

1008 I. A. Alshabeeb, N. G. Ali,S. A. Naser, W. M. Shakir
unsupervised linear method [13]. It works in many steps. The first step is to
mention the number of kclustering and the centroid point. Then we put the
data that are similar to each other or have the lower distance in one group.
Finally, we repeat the steps above to set all data in groups and no object is
moved from cluster to another. The k−means flow chart is shown in figure
2 and the process of the algorithm is demonstrated in figures 3, 4 and 5.
The objective function for k−mean method could be written as:
f=
k
X
i=1
n
X
j=1
(pj −yi)2,(2.1)
where kis the number of clusters, nis the number of cases, pj represents the
point or the object in the cluster and yi is the centroid point for cluster j.
The centroid point represents the mean value for each cluster.
In figure 3, the number of clusters was selected and the centroid point for
each group was found.
In figure 4, the distance between centroid points and the object was com-
puted and assigned the object to the nearest cluster depending on the dis-
tance value. In figure 5, we recalculated the centroid for each cluster and
reassigned objects to groups until there was no change any longer.

A Clustering Algorithm Application in Parkinson ... 1009
Figure 2: Flowchart of k−means algorithm

1010 I. A. Alshabeeb, N. G. Ali,S. A. Naser, W. M. Shakir
Figure 3: k−means at initialization
Figure 4: Centroid and grouping

A Clustering Algorithm Application in Parkinson ... 1011
Figure 5: Recomputed centroid and regrouping
3 Result and Discussion
In this model, the data set was applied on 35 patients at the hospital each
year for 4 successive years. The Xl-miner tool was used to analyze the dataset
and implement the method. The generated results are shown in tables 1, 2,
3 and 4, respectively. In table 1, for the year 2016, the overall age for cluster
size 14 is 75.3 while overall age of cluster size 13 is 63. That means 27 out of
35 of patients are older than 62. The rest of patients are from 44 to 50 years
old. In table 2, for the year 2017, the overall age for cluster size 10 is 62.7
while overall age of cluster size 15 is 52.9. That means 25 out of 35 patients
are older than 50 years and the rest are from 14 to 41 years old. For table 3,
for the year 2018, the overall age for cluster size 11 and size 10 are between
47.7 and 39 years old. The rest of patients are from 15 to 28 years old. In
2019, 24 patients out of 35 between 32 and 48 are suffering from the disease
and 11 patients are from 8 to 23 years old. That means PD seems to affect
younger people compared to 2016 and 2017.

1012 I. A. Alshabeeb, N. G. Ali,S. A. Naser, W. M. Shakir
Table 1: Data summary for 2016
cluster Size age
cluster1 14 75.3
cluster2 13 63
cluster3 3 44
cluster4 5 50.6
Total 35
Table 2: Data summary for 2017
cluster Size age
cluster1 10 62.7
cluster2 15 52.9
cluster3 2 14
cluster4 8 41.75
Total 35
Table 3: Data summary for 2018
cluster Size age
cluster1 11 47.7
cluster2 10 39.6
cluster3 7 15.2
cluster4 7 28.2
Total 35
Table 4: Data summary for 2019
cluster Size age
cluster1 13 48.3
cluster2 11 32.6
cluster3 3 8.3
cluster4 8 23.2
Total 35

A Clustering Algorithm Application in Parkinson ... 1013
References
[1] Fan, Kuan, Pengzhi Hu, Chengyuan Song, Xiong Deng, Jie Wen, Yiming
Liu, Hao Deng. ”Novel Compound Heterozygous PRKN Variants in a
Han-Chinese Family with Early-Onset Parkinsons Disease.” Parkinsons
Disease, (2019).
[2] Priyansha Raj Sinha, Amit Alexander Charan, ”Parkinsons disease: A
review article.” The Pharma Innovation, 6, no. 9, Part H, (2017), 511.
[3] Shianghau Wu, Jiannjong Guo, ”A Data Mining Analysis of the Parkin-
sons Disease,” iBusiness, 3, no. 1, (2011), 71–75.
[4] J. Thomas, R. Theresa Princy, ”Human heart disease prediction sys-
tem using data mining techniques,” In 2016 International Conference
on Circuit, Power and Computing Technologies (ICCPCT), 1-5. IEEE,
(2016).
[5] M. Deepika, K. Kalaiselvi, ”A Empirical study on Disease Diagnosis
using Data Mining Techniques,” In 2018 Second International Con-
ference on Inventive Communication and Computational Technologies
(ICICCT), IEEE, (2018), 615–620.
[6] Xue-Hui Meng, Yi-Xiang Huang, Dong-Ping Rao, Qiu Zhang, Qing Liu,
”Comparison of three data mining models for predicting diabetes or
prediabetes by risk factors,” The Kaohsiung journal of medical sciences,
29, no. 2, (2013), 93–99.
[7] Abolfazl Saghafi, Chris P. Tsokos, Rebecca D. Wooten, ”On Heredity
Factors of Parkinsons Disease: A Parametric and Bayesian Analysis,”
Advances in Parkinson’s Disease, 7, no. 3, (2018), 31–42.
[8] Jiawei Han, Micheline Kamber, Jian Pei, Data mining concepts and
techniques, 3rd edition, Morgan Kaufmann, 2011.
[9] Hina Gulati, P. K. Singh, ”Clustering techniques in data mining: A
comparison,” In 2015 2nd international conference on computing for
sustainable global development (INDIACom), IEEE, (2015), 410–415.
[10] J. Han, M. Kamber. Data Mining: Concepts and Techniques, 2nd Edi-
tion, Elsevier, 2006.

1014 I. A. Alshabeeb, N. G. Ali,S. A. Naser, W. M. Shakir
[11] T. Madhulatha, T. Soni, ”An overview on clustering methods,” arXiv
preprint arXiv:1205.1117, (2012).
[12] Lin Liao, Zhen Jia, Yang Deng, ”Coarse-Graining Method Based on
Hierarchical Clustering on Complex Networks,” Communications and
Network, 11, no. 1, (2019), 21–34.
[13] Ling-Li Jiang, Yu-Xiang Cao, Hua-Kui Yin, Kong-Shu Deng, ”An im-
proved kernel k−means cluster method and its application in fault di-
agnosis of roller bearing,” (2013).
[14] Manyun Lin, Xiangang Zhao, Cunqun Fan, Lizi Xie, Lan Wei, Peng
Guo, ”Polarimetric Meteorological Satellite Data Processing Software
Classification Based on Principal Component Analysis and Improved
k−means Algorithm,” Journal of Geoscience and Environment Protec-
tion, 5, no. 7, (2017), 39.
[15] Siwei Wang, Miaomiao Li, Ning Hu, En Zhu, Jingtao Hu, Xinwang
Liu, Jianping Yin, ”k−means clustering with incomplete data,” IEEE
Access, 7, (2019), 69162-69171.