Högskolan i Skövde

Heart disease is one of the leading causes of mortality, making it crucial to enhance the possibility of diagnosing it at the early stages. This thesis presents interpretability models using SHAP (SHapley Additive exPlanations), PDP (Partial Dependence Plots), and interaction techniques to improve the heart disease classification process. The research indicates that these models are, in general, more accurate and reliable when the results are based on individual and interactive features. SHAP (SHapley Additive exPlanations) provides detailed insights into the importance of each feature compared to other features, whereas PDP (Partial Dependence Plots) explores the effects of features on predictions. 

Analyzing both SHAP and PDP models reveals both similarities and differences in their interpretations. While SHAP offers precise feature importance and interactions, PDP provides a visualization of the relationship between features and the target variable. This comparison helps ensure the consistency and reliability of heart disease prediction. 

In addition, this research examines the consistency of different feature interactions in heart disease prediction by comparing the results from various analytical methods. Using these interpretation methods together gives a clearer understanding of heart disease dynamics, helping society gain a better overall view of the disease and allowing for more focused treatments. The goal of this project is to help healthcare professionals and researchers identify important factors related to heart disease prognosis, improving pre-diagnostic modeling for healthcare systems.