Explainable AI approach for early detection of Parkinson's disease using PPMI Online data.

Accurate and early disease prediction enables patients to plan and improve their quality of life in the future. Early detection of neurodegenerative diseases, such as Parkinson's disease, is a high priority and a significant challenge in which physicians must act quickly to diagnose and predict the risk of disease severity. Machine learning (ML) models combined with feature selection (FS) techniques can assist physicians in quickly diagnosing a disease. FS technique optimally subsets features to improve model performance and reduce the number of tests required for a patient, thereby speeding up diagnosis. This paper proposes an e-diagnosis approach based on ML-FS algorithms to detect Parkinson's disease using data obtained from Parkinson's Progression Markers Initiative (PPMI) Online study. Also, it can be considered patient-oriented research as it uses self-reported online collected data. The results of six FS techniques pre-applied to classification algorithms named logistic regression, random forest, support vector machine, CatBoost, extreme learning machine, and XGBoost are shown in this study. Chi-square, mutual information, and analysis of variance (ANOVA) filter-based FS methods, while sequential feature selection, Boruta, and recursive feature elimination are considered wrapper methods. The outcomes show that random forest when trained on features selected by the recursive feature elimination technique help to build an efficient and effective approach for detecting Parkinson's disease. [ABSTRACT FROM AUTHOR]