Prediction of Parkinson's Disease using Machine Learning and Deep Transfer Learning from different Feature Sets

Abstract

Precursors that appear diminutive hold a world of significance with regard to indicating and diagnosing premature stages of degenerative diseases, especially neurologically concerned ailments like Parkinson's disease (PD). PD usually shows symptoms like spasm in the limbs, jaw or head, rigidness of the limbs and trunk, sluggish movement, etc. and it is crucial to detect Parkinson Disease in early stages by keeping an eye out for these preliminary symptoms. In this paper, various datasets were researched, analyzed and run through certain algorithms to detect several symptoms. The Freezing of Gait dataset was used to predict if there were symptoms related to legs and trunk by analyzing the patient's gait, the Parkinson Clinical speech dataset to detect deviation in audio frequency and lastly the Parkinson Disease wave and spiral drawing dataset which can help find out impairment in writing due to a tremor in hand or arm. The detection of impairment in handwriting seems to be the most convenient method and Convolutional Neural Network using Transfer Learning is implemented on this image dataset.