Voice-Based Classification of Parkinson’s Disease Using Machine Learning: An Extensive Study

Parkinson’s Disease (PD) is a condition observed with neural loss, primarily distinguished by its impact on motor function. In our research, the usage of voice datasets of PD patients and healthy individuals is explored using machine learning (ML) models to classify PD. The dataset is used with the “Synthetic Minority Over-sampling Technique” (SMOTE) algorithm to handle class imbalances and hyperparameter tuning. The dataset is used to train ML models and the accuracies, confusion matrices, and classification reports before and after tuning of the models are observed. DTs are used in this study as they are simple and robust in decision-making problems. RF and XGB are very efficient models that use multiple DTs to give optimal results. RF uses multiple DTs and aggregates their results to yield better accuracy; whereas XGB is a powerful gradient-boosting algorithm that is good at handling missing data. Throughout the study, it is found that the RF model best performs before tuning in terms of classification report characteristics; however, after tuning, its elevated classification report indicates that it is better at handling unseen data after tuning is done.