Vocal features based Parkinson’s detection: An ensemble learning approach

Abstract

Parkinson’s disease (PD) primarily affects the central nervous system. In 2019, over 8.5 million cases were reported, with numbers continuing to rise. This growing prevalence emphasizes the urgent need for early detection and preventive strategies. To resolve this, numerous methods have been introduced, one of them being machine learning technique. By employing deep learning methods on the large-scale datasets, the early prediction and detection of PD is possible. These methods should be precisely evaluated on the basis of vocal features and the best method to predict this neurodegenerative ailment is disclosed. The core objective of this research is to facilitate the medical centers by providing an optimal machine learning technique to early detect PD. In order to decide an ideal method, the renowned machine learning algorithms like Random Forest, K Nearest Neighbor, Naïve Bayes, Gradient Boosting and XGBoost are evaluated according to their performance. Gradient Boosting outperforms earlier results with high recall, low log loss, and overfitting resistance.

Vocal features proved to be valuable indicators for early-stage Parkinson’s detection.

The Gradient Boosting model has scored the highest in terms of all the mentioned parameters, showing a promising result for predicting the occurrence of PD.

Machine learning can play a significant role in supporting clinical diagnosis and decision-making.