Explainable Artificial Intelligence for Parkinson’s disease diagnosis model

Parkinson's disease is characterised by the presence of motor and non-motor symptoms within a complex diagnosis prone to errors. To support this process, Machine Learning models have been developed to distinguish Parkinson's patients from other pathologies with similar symptoms. The model under study uses movement data to detect motor symptoms and questionnaire data to identify non-motor symptoms. These developments enable clinical decision support; however, due to their complexity, interpreting the results can be challenging, which is a crucial factor for doctors and clinicians. This work focused on applying explainable artificial intelligence techniques, specifically, Shapley additive explanations. The results indicate that the most important non-motor symptoms for the model are loss or alteration of the ability to taste or smell, urinary dysfunction, loss of interest, and drooling during the day. Local analysis helped identify patterns in incorrect classifications, especially for patients with Parkinson's disease, while the heterogeneity of the group with different diseases made this analysis challenging. For the model with movement data, no results were obtained due to heterogeneity in feature preprocessing across folds and models.