XAI UTILIZATION AS THE BASIS FOR DEVELOPING A RECOMMENDATION SYSTEM FOR CORONARY HEART DISEASE PREDICTION

• Main Author: Yanggara, Louis
• Format: Final Project
• Language: Indonesia
• Online Access: https://digilib.itb.ac.id/gdl/view/82065
• Institution: Institut Teknologi Bandung

Artificial Intelligence (AI) is essential in fields like banking and healthcare, where it excels at data-driven predictions. However, AI's lack of transparency, fairness, and potential bias in predictions necessitates the use of eXplainable AI (XAI) methods. XAI enhances AI by providing clear explanations for prediction outcomes, crucial in applications such as coronary heart disease (CHD) prediction, a leading global cause of death. This study develops a random forest AI model to predict CHD, achieving an accuracy of 93%. After that, an explanation of the XAI method is given, namely LIME, SHAP, and Anchor which can provide a local explanation of a predicted data instance. The model was trained using pre- processed Framingham Heart Study data so that it has better quality. The results of the XAI testing that has been done show that the LIME method can provide results that are more often chosen by experts. While the SHAP method is less accurate in providing an explanation and the Anchor method provides an explanation that is too general because it produces all features that are considered as anchors. The implemented recommendation system uses the principle of knowledge base system by using expert knowledge related to habits that need to be adapted for each variable. When the user is predicted to have coronary heart disease, recommendations will be given according to the variables that support the prediction of the AI model.