Parkinson’s Detection - Explainable Approach to Temporal Audio Classification

This research addresses the challenge of model interpretability in neural networks, particularly in the medical field for Parkinson's disease (PD) detection. The proposed AI-driven deep learning model leverages vocal deterioration as an early symptom of PD, achieving a high classification accuracy of 90.32% with precise performance metrics. A novel interpretation algorithm is introduced, which analyzes audio files by dividing them into meaningful chunks, assessing each segment's contribution to accurate disease detection. This method not only enhances the reliability and efficiency of PD detection but also ensures scalability and bias-free interpretability of the deep learning model's predictions.