Parkinson disease detection using machine learning with Explainable AI

Abstract

Aside from its effects on the nervous system, Parkinson's disease also modifies the physiological functions that the nervous system regulates. The progressive nature of Parkinson's disease makes it a degenerative disorder. A variety of symptoms, such as rigidity, tremor, speech impediment, sluggishness, and trouble walking, can be intentionally induced. These same principles are applicable to issues like depression, anxiety disorders, and similar ailments. We aim to use machine learning algorithms to identify cases of Parkinson's disease (PD) in patient-created audio recordings. Time Frequency Features, Mel Frequency Cepstral Coefficients (MFCCs), Vocal Fold Features, TWQT features, and Wavelet Transform-based Features are some of the speech signal processing methods that were used to extract clinically important data from the audio recordings in order to evaluate PD. In order to glean useful information from the recordings, these methods were employed. The objective of this research endeavour is to assess a number of these models by means of a number of machine learning techniques, such as Logistic Regression, XGBoost, Adaboost, Decision Tree, Support Vector Machine, Naive Bayes, and Random Forest, among others. The application uses voice attribute-based data preparation, which allows for the measurement of effectiveness. By the end of the experimental evaluation, the XGBoost classifier had achieved the maximum achievable accuracy of 88%. The authors of this study used explainable AI to zero in on the problem of creating models that laypeople might understand.