Comparative Investigation of Classification Algorithms using Parkinson's Disease Acoustic Analysis Dataset to Choose Best Classifier for Best Result

Early diagnostic discovery of Parkinson's Disease (PD) is a serious clinical issue. Owing to the early manifestations of speech impairments in the force of PD patients, the acoustic features of speech have potential to be considered as better early diagnosis standards. In this study, which utilizes a dataset that includes acoustic features associated with PD, we analyze the performance of three machine learning classifiers: Random Forest(RF), Support Vector Machine(SVM), and k-Nearest Neighbors(k-NN) classifiers. In one-second audio segments, a feature extraction was carried out based on principles of jitter, shimmer, harmonics-to-noise ratios, and recursive feature elimination (RFE), as an approach to eliminate the unhelpful features without hindering the model prediction. The classifiers were developed in MATLAB and evaluated based on accuracy, precision, recall, and F1-score using a 10-fold cross-validation method to ensure rigorous assessment. The results showed that RF achieved the highest recall and accuracy (93.0% and 92.3%, respectively), highlighting the suitability of RF as a PD diagnostic model. While k-NN had the lowest results, SVM was precise but scored low on recall. Such works highlight the promise of RF and feature selection in the construction of diagnostic tools for PD, addressing the key concerns of non-invasiveness and improved patient outcomes.