Predictive Models for Chronic Cardiac Disease with LIME and SHAP

With growing machine learning model complexity, there' is a rising need for transparent and interpretable techniques to understand their decision-methodologies frameworks, particularly in vital fields such as healthcare. This paper focuses on applying Local Understandable Model Agnostic Explanations (LIME) and SHapley Additive exPlanation (SHAP) techniques to interpret predictive models for chronic cardiac disease identification, aiming to clarify the model's responses locally and globally. LIME provides local interpretability by offering insights into individual predictions, allowing healthcare professionals to understand circumstances surrounding the decision why a specific decision was made for a particular situation. In contrast, SHAP delivers global interpretability by explaining the overall behaviour of the model and its key predictive features across the entire dataset. Through extensive experimentation on clinical datasets, we evaluate the effectiveness of these methods in providing actionable insights to healthcare professionals, facilitating trust and understanding of the predictive models. Our findings contribute to advancing interpretability research in healthcare and offer valuable guidance for employing these techniques in clinical decision support systems for chronic cardiac disease management.