A Comprehensive Voice Data Analysis for Parkinson's Disease Prediction via Machine Learning Techniques

This work aims to develop a Machine Learning-based system to classify patients with Parkinson's disease. The existing systems have limitations, including limited generalizability and issues with imbalanced datasets. As such, there is a need for a more robust and efficient machine learning approach to accurately differentiate between individuals with Parkinson's disease and those who are healthy. Since speech disorder is a symptom of the disease, we have used the Disease Voice Dataset from the UCI database in this research, which has 195 memos of the voice of 31 people, of which 23 are affected by the disease. We have preprocessed and balanced the data using Min-Max Normalization and Synthetic Minority Oversampling Technique. Decision Tree, K-Nearest Neighbours and Support Vector Machine classifiers are trained on voice, followed by hyperparameter tuning using GridSearchCV that improves the performance of the models significantly. We implement five parameters: accuracy, precision, recall, F1-score, and AUC-ROC value. Among the used algorithms, the highest classification accuracy obtained is 97.44% using SVC CLASSIFIER on the balanced dataset. The implemented work surpasses the limitations of previous approaches and provides a more accurate and efficient paradigm for the benchmarked detection of Parkinson's disease.