AN EXPLAINABLE MACHINE LEARNING FRAMEWORK WITH ROBUST CROSS-VALIDATION FOR PARKINSON'S DISEASE PREDICTION USING VOICE BIOMARKERS

Abstract

A chronic neurological condition that impairs mobility is Parkinson's disease. Slower motions, tremors, and stiffness might be symptoms of Parkinson's disease. Although a precise diagnostic test does not yet exist, machine learning can be used to estimate a person's likelihood of having Parkinson's disease based on particular biomarkers. The purpose of this article is to forecast Parkinson's disease using machine learning algorithms.

Early detection may be delayed by traditional clinical diagnosis, which mostly relies on medical knowledge and symptom monitoring. Thus, machine learning (ML)-based automated prediction systems can greatly enhance early diagnosis and assist medical professionals.

This work uses clinical and biological speech datasets to develop a machine learning-based classification model for Parkinson's disease identification. XGBoost (Extreme Gradient Boosting), Random Forest, and Support Vector Machine (SVM) are three potent supervised learning techniques used by the system. Data preparation procedures, such as feature scaling, normalisation, cleaning, and feature selection, are carried out to raise the dataset's quality and boost model performance. In order to properly evaluate the dataset, it is then separated into training and testing sets. Metrics including F1-score, confusion matrix, recall, accuracy, and precision are used to assess the models' performance. In general, XGBoost and Random Forest offer more accuracy and resilience, according to comparative studies, whereas SVM handles complicated categorisation boundaries well.