A Review of Machine Learning Used in the Diagnosis of Parkinson’s Disease

Abstract

Parkinson’s Disease (PD) is projected to impact an increasing number of individuals due to the anticipated growth of the global elderly population. While there is currently no cure, early diagnosis remains crucial for extending the quality of life for individuals with PD. Machine Learning (ML) techniques have been found to be effective in facilitating remote monitoring and enabling early diagnosis of PD. ML algorithms have shown to be able to achieve higher accuracy diagnostics compared to experts, and there is still room for improvement. This paper aims to provide a comprehensive overview of recent developments in diagnosing PD using ML. The study investigates eight of the most widely used ML algorithms, namely Support Vector Machines (SVMs), Neural Networks (NNs), Ensemble Learning, K Nearest Neighbours, Logistic Regression, Decision Trees, Naive Bayes and Discriminant Analysis, to provide a thorough analysis of their applicability and effectiveness in PD diagnosis. This paper will focus on these algorithms as they are the basis of many other variants, and they are most popularly researched and used. The paper discusses the strengths and weaknesses of each algorithm, presents examples of their usage, and highlights their efficacy with different PD indicators. Moreover, this paper reviews some of the most influential works in recent years, identifying the most significant challenges in the field of PD diagnosis. It highlights how researchers have attempted to address them and outlines directions for future research. First, this paper reviews the ML techniques used in diagnosis of PD. Then, we discuss the ML models’ shortcomings and strength. Finally, we discuss the challenges and future directions in research of this field. Notably, the study shows that SVMs and NNs emerge as popular choices due to their efficacy with commonly used datasets in PD diagnosis.