AN INTERPRETABLE SPEECH ANALYTICS FRAMEWORK FOR EARLY PARKINSON'S DISEASE IDENTIFICATION USING K-NEAREST NEIGHBORS

Abstract

Parkinson's disease (PD) is a long-term, degenerative neurological condition that affects the nerve system of the body and causes problems with speech, movement, and coordination. Dysphonia, a voice problem marked by changes in vocal quality, pitch, and loudness, is one of the early signs of Parkinson's disease. Vocal feature analysis becomes a useful method for early identification because between 70 and 90 percent of people with Parkinson's disease have speech problems. Previous research has used a variety of machine learning models to identify Parkinson's disease (PD) using speech signals; nevertheless, issues such class imbalance, optimal feature selection, and limited interpretability are still common. In order to get over these restrictions, this study presents a prediction framework for Parkinson's disease identification based on speech data that makes use of the K-Nearest Neighbours (KNN) algorithm. KNN, a non-parametric and instance-based learning method, classifies patients by comparing their voice feature patterns to those of their nearest neighbours in the dataset. Its simplicity, versatility in dealing with non-linear data distributions, and effectiveness with small to medium-sized datasets make it appropriate for medical diagnostic applications. The algorithm can achieve high accuracy in differentiating PD patients from healthy persons by optimising the value of k and employing distance measures such as the Euclidean or Manhattan distance. Furthermore, feature normalisation and dimensionality reduction techniques are used to enhance KNN's performance and dependability. This strategy attempts to improve the accuracy of early Parkinson's identification while maintaining interpretability, providing a clinically meaningful and data-driven alternative to existing diagnostic procedures.