Machine Learning-Based Detection of Parkinson’s Disease: A Comparative Study of Classifiers on Clinical Voice Data

Abstract

: Parkinson’s disease is a long term , progressive neurological disorder that gradually impairs motor functions such as involuntary tremors, rigidity, slow movements, and challenges with maintaining balance . These symptoms significantly impact the quality of life , by making early and accurate diagnosis important for timely intervention and effective management. Unfortunately, traditional methods for diagnosis-including neurological examinations , neuroimaging , and clinical assessments-are often resource-intensive, costly, and may not be accessible in many healthcare settings. Moreover , these methods can sometimes delay diagnosis until the disease has advanced , which diminishes treatment effectiveness . In light of these challenges , recent scientific efforts have focused on developing more accessible and non-invasive screening tools. Among the most promising avenues is the analysis of speech and voice signatures. Vocal impairment is one of the earliest symptoms in many PD patients, resulting from neuro degeneration affecting the circuits controlling voice production. These vocal impairments include subtle changes in pitch modulation, increased jitter and shimmer, and a loss of harmonic structure in speech. Utilizing speech signal processing techniques, researchers can extract detailed acoustic features that reflect these underlying motor deficits. These features provide a valuable opportunity for early detection, even before more apparent physical symptoms manifest , which is vital for initiating treatments that can slow disease progression. This project investigates the efficiency of using machine learning algorithms in classifying Parkinson’s disease based on voice recordings. Specifically, we compare four prominent classifiers which include Support Vector Machine , Random Forest, Logistic Regression, and K-Nearest Neighbors - to determine which provides the most reliable early diagnosis . Using a well-established dataset of clinical voice samples, we carefully extracted relevant acoustic features related to pitch , frequency , amplitude , and other paralinguistic markers indicative of vocal impairment . These features were then used to train and test each classifier under consistent experimental protocols. The models’ performance was measured through classification accuracy and confusion matrix analysis, which reveal both the overall effectiveness and the error distribution of each approach. Our results demonstrate that the Support Vector Machine achieved the highest accuracy of 87.18 % , outperforming the other classifiers. Random Forest and Logistic Regression yielded comparable results, both reaching about 82 % , while KNN lagged slightly behind with 76.92 %. Importantly, all models kept false positive and false negative rates relatively low, confirming their potential for practical use in clinical settings. The findings from this project highlight the importance of using voice feature analysis along with machine learning as an effective, cost-efficient , and non-invasive method for the early detection of Parkinson’s disease.