KNOWLEDGE-BASED CREDIT GRANTING SYSTEM IN INDONESIA WITH ARTIFICIAL INTELLIGENCE AND XAI APPROACH

• Main Author: Steven Supardi, Rolland
• Format: Final Project
• Language: Indonesia
• Online Access: https://digilib.itb.ac.id/gdl/view/73248
• Institution: Institut Teknologi Bandung

Knowledge-based systems with a blackbox model generally lack internal information as explanations for decisions. This is not suitable for high-risk fields such as credit, both in terms of user needs and regulations (Personal Data Protection (PDP) and General Data Protection Regulation (GDPR)). Therefore, XAI is commonly used for explanations in such systems. However, its application in credit granting has not been found with Indonesian data yet, as the regulations are still new. This final project aims to create a knowledge-based credit granting system in Indonesia with local explanations. This can be achieved by using Indonesian knowledge base, searching for machine learning algorithms and XAI methods with a level of explanation that can be easily understood by users. To achieve this, the final system created is the best solution based on a comparison of Decision Tree and XGBoost models with LIME, SHAP, and Anchors methods. Evaluation metrics such as accuracy and feedback from domain experts are used to find the best system. The data used in the system needs to be adapted for machine learning algorithms and XAI to provide sufficiently accurate predictions with easily understandable explanations. The test results show that the Decision Tree explanation method provides the best results compared to other tested methods. However, the system still has weaknesses in terms of accuracy and explanations of important features, which are still too simplistic and dominated by certain features.