Comparison Of Classification Methods To Detect The Parkinson Disease

 

               Comparison Of Classification Methods To Detect The Parkinson Disease

Parkinson's disease is a chronic neurological degenerative disease affecting the central nervous system responsible for essentially progressive evolution movement disorders. The detection of this disease is made using a clinical diagnosis made by an expert. To save time and for more comfort in the Diagnostic and also to increase the efficiency of treatment through preventive detection, classification has found its place in the detection of this disease. For this purpose, many classification algorithms were used to achieve the best results, but the problem is which classifiers may be the most efficient for this detection, because each classification algorithms was applied to a local database, what influences the results. In this paper we have tried to apply three types of classifiers, k-near neighbor "k-NN", the Naive Bayes "NB" and support vector machines "SVM", on the same database to compare and to know which of the three classifiers will be the most efficient.

Parkinson's disease is in second place behind the Alzheimer regarding the most common neurological diseases.

 

It is a chronic neurodegenerative disorder, slowly progressive, usually of unknown origin. It affects a structure of a few millimeters at the base of the brain and which is composed of dopaminergic neurons that are gradually disappearing. Their function is to produce and release dopamine, a neurotransmitter essential to the control of body movements, particularly the automatic movements.

 

Parkinson's disease begins 5–10 years before the onset of clinical symptoms, when about half of dopaminergic neurons disappeared. The diagnosis can be easy due to the presence of at least two of the following three symptoms:

 

slowness of movement (bradykinesia)

 

a resting tremor of the hand and/or foot-sided

 

stiffness (hypertonia)

 

The diagnosis can be very difficult by the existence of very different non-typical signs, such as depression, pain, and fatigue. So, each patient has special signs compared to others patients.

 

This disease mainly affects those over 60 years, but among patients suffering from this disease there is 10% less than 50 years.

 

Clinical diagnosis for Parkinson's disease is subjective because it is based on signs and symptoms appeared in the patient, or we said that every patient has specific signs, so there is no definitive test for detection of this disease. For this we need new methods able to detect this disease permanently in its early stages, in order to benefit the maximum possible of medical treatment to increase its efficiency. For this purpose, the automatic classification based on learning classification algorithms is very interesting.

 

Because Parkinson's disease affects the patient's voice, much studies has been done on the classification of patients based on the voice and speech. This classification arises on the extraction of the voice features to detect anomalies compared to voice features from a healthy person, in order to predict Parkinson disease.

 

In this work, we will try to apply three different types of classifiers on the same database, in order to determine the characteristics and judge on which classifier is more effective in the detection of Parkinson's disease from voice.

 

Then, we selected three classifiers among the most used classifiers, k-nearest neighbor, naive Bayes and support vector machines to classify voice signals into two classes “healthy people and patients”.

 

In the following section, we present the different methods of diagnostics for Parkinson's disease, in section III, we will give the description of our process, Section IV, presentation of the results, comparison and choice of the most efficient classifier. Then we conclude our paper in section V, and highlight the direction for future research.