Voice-Based Detection of Parkinson’s Disease: A Multi-Machine Learning Model Approach

This study explores the voice-based detection of Parkinson’s disease (PD), a progressive neurological disorder characterized by motor impairments. Utilizing a dataset of 195 voice recordings from 31 individuals, including 23 diagnosed with PD, we analyzed features such as fundamental frequency variations, amplitude variations, and nonlinear dynamical complexity measures. This study introduces a novel approach by employing a combination of feature extraction and traditional machine learning algorithms for voice-based PD detection. The dataset was divided into training, validation, and testing sets using a 70/15/15 split strategy. Preprocessing included inactive, univariate, and multivariate analysis, along with correlation heatmap generation. Machine learning models, including Decision Tree, K-Nearest Neighbors, Random Forest, XGBoost, and Gradient Boost, were employed for classification. Model evaluation was performed using crossvalidation and metrics such as accuracy, precision, recall, F1score, sensitivity, and specificity. Gradient Boost achieved the highest accuracy (97%) and F1-score (0.97), while Random Forest demonstrated excellent sensitivity (95%) and specificity $(100 \%)$. While deep learning models have shown promise, this study demonstrates the effectiveness of traditional machine learning algorithms for PD detection. Future work will explore the integration of multimodal data to enhance accuracy. These results highlight the potential of machine learning for early and accurate PD detection through non-invasive voice analysis, paving the way for advanced diagnostic tools in healthcare..