A COMPARISON OF FEATURE SELECTION METHODS FOR DIAGNOSIS OF PARKINSON’S DISEASE FROM VOCAL MEASUREMENTS

41551

Parkinson’s disease is a neurological disorder which affects the quality of life of the patients and has major social and economical impact. To diagnose the disease, clinical examinations and observations with Unified Parkinson’s disease Rating Scale (UPDRS) have been used, but especially on the initial phase of the disease, this method may be insufficient. In this paper, we used biomedical voice measurements obtained from sustained phonation samples for detection of Parkinson’s disease. We compared the six different types of feature selection procedures. These are Bhattacharyya, information gain, relief, minimum-redundancy maximum-relevancy (MRMR), t-test, and support vector machine methods based on recursive feature elimination (SVM-RFE). It was found that SVM-RFE gave the best recognition results with the 95.13% classification accuracy for Parkinson’s disease dataset.

402

Parkinson’s disease , classification , support vector machine , feature selection , SVM , RFE

Sigma Journal Of Engineering and Natural Sciences

414

Sigma Journal Of Engineering and Natural Sciences