A Comparative Study of Machine Learning Models for Parkinson’s Disease Detection

Abstract

Parkinson’s disease is a major public health concern, affecting over 6 million people worldwide. The objective of this paper is to assist doctors and clinicians in accurately detecting the disease at an early stage. Previous research proposed various models that gave very high accuracy. However, very few of them examined the processing time of each model, which is an important consideration in decision making. The most common method for diagnosing this disease is through voice signal recordings. This paper formulates 10 machine learning-based predictive models on a biomedical voice measurement dataset. A genetic algorithm is applied as a feature selection algorithm. The highest prediction accuracy after running 10 generations is 97.96%. The features of the most accurate model are reduced from 22 to 9 features. The processing time of the most accurate model is 1.83 seconds. The best improvement in accuracy after applying this feature selection algorithm is 16.33%.