Parkinson’s disease Detection With SVM classifier and Relief-F Features Selection Algorithm





 1 Cette adresse e-mail est protégée contre les robots spammeurs. Vous devez activer le JavaScript pour la visualiser.,';document.getElementById('cloak1f70e2da465687f9d699e5dfa1f50a2d').innerHTML += ''+addy_text1f70e2da465687f9d699e5dfa1f50a2d+'<\/a>'; Cette adresse e-mail est protégée contre les robots spammeurs. Vous devez activer le JavaScript pour la visualiser.r, 3 Cette adresse e-mail est protégée contre les robots spammeurs. Vous devez activer le JavaScript pour la visualiser.,';document.getElementById('cloak8488bd8446181625ef675e1b91d0dc43').innerHTML += ''+addy_text8488bd8446181625ef675e1b91d0dc43+'<\/a>';

4Cette adresse e-mail est protégée contre les robots spammeurs. Vous devez activer le JavaScript pour la visualiser.,';document.getElementById('cloak7d1f1d1ca93995d7fab0dabd2b0976f8').innerHTML += ''+addy_text7d1f1d1ca93995d7fab0dabd2b0976f8+'<\/a>';

Biomedical Engineering Laboratory

Tlemcen University, Algeria



Abstract—Artificial intelligence techniques have been extensively used for the identification of several disorders related with the voice signal analysis, such as Parkinson’s disease (PD). Parkinson’s disease is a neurodegenerative disorder with a long time course and a significant prevalence, which increase significantly   with   age.   Although   the   etiology   is   currently unknown, the disease presents with neurodegeneration of regions of the basal ganglia. The onset occurs later in life, and the disease is diagnosed clinically, requiring the identification of several factors such as distal resting tremor, rigidity, and bradykinesia. The common thread throughout the range of symptoms is motor dysfunction, and recent reports have focused on dysphonia, the impairment in voice production as a diagnostic measure. In this paper, we propose a features selection algorithm to reduce features’ number  and increase the performances; the feature set is analyzed and reduced from 22 to 10. Then an automatic recognition system is applied using Support Vector Machine (SVM) classifier and 10-fold cross validation method, experimental results show that the performance of relief-F is comparable with other method, with a correct rate of 96, 88% to distinguish the healthy people from those with Parkinson’s disease.


Keywords-       Artificial   intelligence,   SVM,   features   selection,

Parkinson’s disease, Vocal records, Dysphonia.



The use of classifier systems in medical diagnosis is increasing gradually. Recent advances in the field of artificial intelligence have led to the emergence of expert systems and Decision Support  Systems  (DSS)  for  medical  applications. Moreover, in the last few decades computational tools have been  designed  to  improve  the  experiences and  abilities  of doctors  and  medical  specialists  in  making  decisions  about their patients. Without doubt the evaluation of data taken from patients and decisions of experts are still the most important factors in diagnosis. However, expert systems and different Artificial Intelligence (AI) techniques for classification have the potential of being good supportive tools for the expert. Classification systems can help in increasing accuracy and reliability of diagnoses and minimizing possible errors, as well as making the diagnoses more time efficient [1].


Parkinson’s  disease  is  a  degenerative  disease  of  the nervous system, which as progresses, causes the patients to have  difficulty in  walking, talking, thinking or  completing other simple tasks [2-4]. PD usually affects people over the age  of  50  and  for  most  elderly  people  with  Parkinson’s

disease, physical visits to the clinic for diagnosis, monitoring, and treatment are difficult [5].


The main symptoms of PD are tremor, rigidity, and other general movement disorders. Of particular importance to this study, vocal impairment is also common [6], [7], with studies reporting 70%–90% prevalence after the onset of the disease [7]–[9]. In addition, it may be one of the earliest indicators of the disease [10], [11], and 29% of patients consider it one of their greatest hindrances [9].


[12] Presented a collection of feature selection algorithms to deal with PD recognition, and Little& al. [2] developed a remarkable study about PD identification and also introduced a new feature to improve the effectiveness of such systems. Revett & al. [13] introduced the Rough sets theory for feature selection in the context of PD automatic recognition. Finally, Guo et al. [14] have introduced a learning function composed by a Gaussian Mixture Model and Genetic Programming in order to accomplish this task.


Several works have addressed the PD automatic identification.  Jervis  et   al.   [15]   for   instance,  evaluated Artificial Neural Networks with Multilayer Perceptron (ANN- MLP) to distinguish PD from Huntington’s disease and Schizophrenic patients. Cutsuridis & al. [16] also applied ANN-MLP for PD bradykinesia recognition, which is characterized by speech impediment, and is also known by akinesia [17]. [18] has made a comparative study of multiple classification algorithms on the same data set used in this study  with  regard  to  neural  networks, DMNeural analysis, regression  analysis  and  decision  trees  with  the  presented results of classification accuracy of 92.9%, 84.3%, 88.6% and

84.3%, respectively.

 Another study published by [19] on the imbalanced problem in biomedical data, uses a sampling scheme in collaboration with a Naive Bayes classifier to deal with the imbalanced problem. The sampling pattern is to start with a small portion of the data to train the classifier, and then successively to increase the number of training samples regardless of the initial class distribution. The method shows promising results with positive predictive rates of 66.2% for normal subjects and 90.0% for subjects with PD.

 In this study, SVM classification with feature selection

was used to diagnose the Parkinson ’s disease. The data set

was taken from UCI machine learning repository. (,  last accessed: May 2012).   It was observed that the proposed method yielded the highest classification accuracy (98.53%) for a subset that contained ten features. Also, other measures such as sensitivity, specificity, were used to show the performance of SVM with feature selection.



télécharger l'article