Optimizing Parkinson’s Diseases Detection from Vowel Phonation: A Bayesian Approach with K-Nearest Neighbours

This study explores the optimization of Parkinson’s Diseases (PD) detection utilizing vowel phonation data through a Bayesian approach coupled with the K-nearest neighbours (KNN) algorithm. Using a variety of speech signal processing techniques, 754 characteristics were retrieved from sustained 'a' vowel phonation recording from 252 participants (188 PD patients and 64 healthy controls) using an open-source dataset from the University of California Irvine. Each subject contributed three recordings, yielding 756 instances in the dataset. Employing Bayesian optimization, we fine-tuned the parameters of the KNN model and achieved promising validation and test accuracies of 92.6% and 93.3%, respectively. The model's effectiveness was also confirmed by the area under the receiver operating characteristic curve (AUC), which had values of 0.9543 and 0.9685 for the test and validation sets, respectively. Notably, the minimum classification error obtained was 0.073727, demonstrating the robustness of the optimized model. The optimal configuration of the KNN model involves setting the number of neighbours to one and employing the Spearman distance metric with inverse distance weighting. These results highlight how Bayesian optimization might improve the ability of machine learning models to diagnose Parkinson's disease, especially when it comes to vowel phonation analysis.