Multi-Model Machine Learning Techniques for Effective Parkinson's Disease Prediction and Diagnosis

It may be difficult to make an accurate diagnosis of Parkinson's disease (PD) in its early stages due to the emergence of minute variations. Parkinson's disease (PD) is a neurodegenerative condition that impacts the neurological, physiological, and behavioral systems of the brain. Slow movement, known as bradykinesia, is a prevalent sign of this illness. Symptoms start to show up in midlife and get worse over time. A change in speech patterns is one of the earliest indicators of Parkinson's disease. In this study, it was proposed to diagnose subjective diseases using supervised classification techniques such k-nearest neighbor (K-NN), artificial neural network (ANN), support vector machine (SVM), and naïve Bayes (NB). To increase robustness beyond what a single model might provide, we have employed a method known as ensembling. Ensemble learning trains numerous models and then integrates their predictions, as opposed to relying solely on one model. This paper investigates four distinct ensemble algorithms: Adaboost, bagging, majority voting, and weighted voting. A dataset of gathered voice sounds will be used to test the proposed methods. There are 195 biological sounds in the sample, including 147 phonetics for PD patients and 48 for healthy people. The analysis found that an ensemble strategy utilizing decision tree classifiers, MLP, SVM, and the majority voting technique yielded superior results. Specifically, with an accuracy rate of 96.3%, the ensemble approach fared better than the individual classifiers.