Comparison of Machine learning models for Parkinson's Disease prediction

Abstract

Parkinson's Disease (PD) is a chronic degenerative disease that mainly affects the nervous system and motor controls in human beings. Early symptoms such as muscle stiffness, tremors, impaired balance and difficulty with walking are considerably less noticeable. Blood tests and Scans also do not provide sufficient evidence for early diagnosis. Hence it is very difficult for doctors to diagnose the onset of Parkinson's Disease. However, smearing of speech gives an early warning and can be effectively used for the prediction of PD. This paper, the voice recording samples of Parkinson's disease affected and healthy patients have been used for PD prediction. Thirteen predictive models using various Machine Learning techniques have been formulated using the University of California, Irvine (UCI) dataset. A comparative study of these predictive models has been carried out on the UCI dataset consisting of biomedical voice recording samples of healthy and Parkinson's Disease affected peoples. These predictive models have been trained and tested for their accuracy and efficiency. The performance analysis of the best five models has been presented in this paper, for accurate prediction of Parkinson's Disease at an early stage. The processing speed of these models has also been analysed, to assess their suitability for light weight mobile applications in the ubiquitous computing environment. © 2020 IEEE.