Detecting Parkinson's disease from Speech signals using Boosting Ensemble Techniques

Abstract

The most economical technique for diagnosing Parkinson's disease is acoustic analysis of human voice which is a non-intrusive, dependable, and simple. The first sign of Parkinson's disease is a voice change from normal. The complexity of sustained speech signals produced by normal speakers and Parkinson's disease patients is described using nonlinear dynamic approaches. The use of such algorithms will have a good influence on the development of an e-healthcare system for patients, allowing for a faster treatment procedure and a considerable reduction in illness severity. Several performance indicators, including F1 Score, recall, accuracy, and precision of classifiers like Adaptive Boost, Gradient Boost, Light Gradient Boost, and XGradient Boost have all been assessed. 30% of the dataset is used for testing, while 70% is for training. The best was discovered to be XGradient, which has 87.39% accuracy rate. A feature significance analysis was also used to discover key characteristics for categorizing Parkinson's patients.