CLASSIFYING PARKINSON’S DISEASE FROM PATIENT’S ACOUSTIC FEATURES USING DEEP LEARNING

Abstract

Parkinson’s Disease (PD) refers to a neurological disorder that is caused due to damage to a part of the brain called the substantia nigra; besides this, the diagnosis of PD is so much more costly and lengthy process. Hence, a cost-effective and efficient system will be helpful for PD patients. Nowadays, the advancement of the algorithms of artificial intelligence is an opportunity to develop an efficient diagnosing system. Till now, no permanent cure has been established; however, early diagnosis of PD can be helpful to lead a better life. According to medical science, speech problem is one of the essential symptoms of PD. Therefore, 22 vocal features of the UCI machine learn ing repository speech dataset are investigated for diagnosing PD patients. Moreover, Artificial Neural Network (ANN), Ensemble Learning (EL) and Machine Learning (ML) are utilized for PD classification. As we need a faster classification system, feature selection techniques have great importance in developing a better system; we investigate excluding and including feature selection. Low Variance Filter, Analysis of Variance, Principle Component Analysis and Linear Discriminant Analysis, feature se lection techniques, are incorporated to select the vital feature. As a result, our proposed ANN model achieves 100% accuracy, F1 score of 100%, recall of 100%, precision of 100%, Kappa Score of 1, and AUC of 1. The efficiency of our proposed on different datasets is demonstrated by comparing with recent research works. At last, an accuracy and classification time trade-off is established to determine the best classifier.