HYBRID MODEL FOR PREDICTING PARKINSON'S DISEASE FROM SPEECH AND HANDWRITING MODALITIES

Abstract

Parkinson's disease is a central nervous system disorder that affects the movement of an individual. it has been observed that patients with parkinson's disease suffers from handwriting abnormalities, stooped posture, speech or voice disorders etc. The work was intended to implement a generalized machine learning model capable of predicting pd from the early stage symptoms. in this study, the speech datasets from uci machine learning repository and spiral and waves images from handwriting datasets were experimented to study the accuracy of the combination model. In order to improve the accuracy of prediction, the features extracted from the speech datasets were jitter,shimmer,nhr,dfa and ppe. also, the features extracted from handwriting datasets were pressure,grip angle,timestamp, radial velocity,speed etc. Different machine learning models like cnn, lstm,resnet etc were experimented on the above datasets. From the study, it was observed that a cnn/lstm model with proper hyper parameter tuning had worked well compared to other models being used in this work. the accuracy of cnn/lstm on the speech dataset was 88% and on the hand writing dataset was 92%..