A Hybrid Feature Selection and Machine Learning Approach for Parkinson’s Disease Detection from Voice Signals in IoT-Enabled 6G Networks

Abstract

With the rapid advancement of the Internet of Things (IoT) and Sixth Generation (6G) networks, real-time health monitoring has become increasingly effective, facilitating continuous tracking of neurodegenerative diseases such as Parkinson’s disease (PD). Voice signal analysis serves as a crucial diagnostic tool for PD detection; however, the high dimensionality of extracted features poses significant challenges for machine learning classification. Traditional feature selection methods often rely on generic fitness functions, which may not optimize classifier performance. The proposed adaptive Particle Swarm Optimization (PSO) method uses specific fitness functions for each classifier while increasing its performance as a feature selection technique. The proposed method develops dynamic subset evaluations through classifier accuracy metrics instead of conventional information gain and entropy fitness measurements to achieve optimal selections suitable for different learning algorithms. By implementing this innovative technique, classification performance achieves higher levels while retaining lower complexity. The Random Forest (RF) classifier, together with PSO-based feature selection, led to a significant accuracy enhancement of 98.3% on a Parkinson’s disease voice dataset, starting from 93.22% without feature selection. The proposed system brings an operationally efficient method to power IoT and 6G healthcare through real-time correct PD detection. The adaptive fitness approach of PSO enables better results than traditional feature selection methods when operating under the resource limitations commonly known in IoT systems.