Explainable AI-Based Clinical Decision Support System for Obesity Comorbidity Analysis

This paper presents a novel Clinical Decision Support System based on eXplainable Artificial Intelligence (XAICDSS) as a comprehensive structured tool consisting of three main parts: predictive models, XAI interpretation, and a graph-based visualization of non-communicable pathologies. Machine learning models are proposed to predict the risk factors related to the direct association between obesity and comorbidities such as cardiovascular, heart disease, and diabetes. Multilayer perceptron and extreme gradient boosting are chosen among different machine learning algorithms as the best performing for the risk factors prediction of the selected comorbidities. They perform prediction with an accuracy of 0.72 for diabetes, and 0.73 for cardiovascular and heart disease. The intuitive XAI interface gives the end-user insight into the machine learning decision process, while the graph visualization links such co-occurrent pathologies to many other non-communicable diseases and provides a global view to healthcare professionals, usable for obesity and associated pathologies prevention and long-term treatment and care. © 2023 IEEE