Design of an Efficient Prediction Model for Early Parkinson’s Disease Diagnosis

Abstract

Parkinson’s Disease (PD) is a long-lasting and progressive brain disorder that disrupts the body’s nervous system pathways. This disruption leads to various issues with movement and control, leading to various symptoms, including tremors, stiffness, and difficulty with movement and coordination. In the early stages of this condition, the patients struggle to speak and also speak slowly. Dysphonia, a speech impairment or alteration in speech, is experienced by 70 to 90 percent of Parkinson’s patients and is an early indication of the disease. Hence, speech can be a vital modality for an early stage of PD diagnosis. In literature, various Machine Learning models are implemented for PD diagnosis based on speech data. However, issues like class imbalance, feature selection, and interpretable prediction analysis are not addressed effectively. Moreover, the accuracy and trustworthiness of the prediction results are essential for providing better healthcare services. Thus, we propose an Interpretable Feature Ranking XGBoost (IFRX) model that effectively addresses the above-mentioned issues. The proposed model has a sequence of processes, such as data preprocessing, class balancing, feature selection, classification, and eXplainable Artificial Intelligence (XAI). We trained the IFRX model based on speech data, which ranks the relevant features, builds an XGBoost classifier and ranked the features according to their relevance in diagnosing PD using Shapley Additive exPlanations (SHAP). Using the proposed model, we implemented eight Machine Learning classifiers for PD diagnosis based on speech data. Among these classifiers, the XGBoost approach shows better prediction performance with an accuracy of 96.61%.

Keywords

• Parkinson’s disease
• feature selection
• XGBoost
• explainable artificial intelligence
• machine learning