Utilizing Explainable AI for Interpreting Machine Learning Model Results in Ceria Credit Scoring

This study aims to improve the transparency of machine learning models in credit scoring using various Explainable Artificial Intelligence (XAI) methods. The methods used include SHAP, BRCG, ALE, Anchor, and ProtoDash to explain the prediction results of machine learning models, namely logistic regression, XGBoost, and random forest. This study applies a quantitative approach with a comparative method, where Ceria loan application data from Bank Rakyat Indonesia (BRI) is analyzed using a machine learning model, then evaluated using the Explanation Consistency Framework (ECF). The results show that the XAI method can improve understanding of model decisions, with SHAP and ALE effective for global explanations, while Anchor and ProtoDash provide in-depth insights at the individual level. Evaluation using ECF shows that the post-hoc method has high consistency, although Anchor has limitations in the aspect of axiom identity. In conclusion, the XAI method can help improve trust and transparency in credit scoring at BRI. Keywords: Explainable Artificial Intelligence; Credit Scoring; Machine Learning; Model Interpretability; Explanation Consistency Framework Abstrak Penelitian ini bertujuan untuk meningkatkan transparansi model pembelajaran mesin dalam penilaian kredit menggunakan berbagai metode Explainable Artificial Intelligence (XAI). Metode yang digunakan antara lain SHAP, BRCG, ALE, Anchor, dan ProtoDash untuk menjelaskan hasil prediksi model pembelajaran mesin yaitu regresi logistik, XGBoost, dan random forest. Penelitian ini menggunakan pendekatan kuantitatif dengan metode komparatif, dimana data pengajuan pinjaman Ceria dari Bank Rakyat Indonesia (BRI) dianalisis menggunakan model machine learning , kemudian dievaluasi menggunakan Explanation Consistency Framework (ECF). Hasilnya menunjukkan bahwa metode XAI dapat meningkatkan pemahaman keputusan model, dengan SHAP dan ALE efektif untuk penjelasan global, sementara Anchor dan ProtoDash memberikan wawasan mendalam pada tingkat individu. Evaluasi menggunakan ECF menunjukkan bahwa metode post-hoc memiliki konsistensi yang tinggi, meskipun Anchor memiliki keterbatasan pada aspek identitas aksioma. Kesimpulannya, metode XAI dapat membantu meningkatkan kepercayaan dan transparansi dalam credit scoring di BRI. Kata Kunci: Explainable Artificial Intelligence; Credit Scoring; Machine Learning; Model Interpretability; Explanation Consistency Framework