Parkinson's Disease Progression Prediction Using Transformer-based Time-Series Models and Explainable AI (XAI)

This study investigates the application of an Optimized Explainable AI Framework (OEAIF-PD) for the detection and progression monitoring of Parkinson's Disease (PD). The framework uses Machine Learning (ML) to simulate the real-world health care scenarios and forecast the clinical progression considering the Unified Parkinson's Disease Rating Scale (UPDRS) scores of patients, their medication state and synthetic observation data. Besides, feature engineering alongside sliding window helps in handling issues in time-series data that needs preprocessing. In the OEAIF-PD framework, the interpretable model is based on SHapley Additive exPlanations (SHAP) to improve the model’s outcome for confidence by clinicians. In evaluation, the current model, the OEAIF-PD is found to deliver better prediction accuracy that enhances from traditional models through demonstrating lower Mean Absolute Error (MAE) and Root Mean Square Error (RMSE) than the current models. The framework’s stability is confirmed with the residual analysis, and SHAP results reveal major feature importance as the interpretation is improved. The findings in this work illustrate the ability of OEAIF-PD to enhance diagnosis, track PD progression, and confirming Parkinson’s disease, while overcoming the drawbacks of traditional approaches. This approach is best suited to healthcare uses and could easily be replicated in other areas of medicine, presenting a solution that is easily scalable and fully transparent.