Rural Bank Credit Prediction in Indonesia Using K-Means Clustering And Ensembled Machine Learning Models

In Indonesia, the scarcity of traditional banking facilities in rural areas has led to the establishment of Bank Perkreditan Rakyat (BPR), or rural banks, which aim to provide credit to underserved communities and foster economic development and financial inclusion. The ability to predict rural bank credit in these areas is of paramount importance, as it enhances the banks' ability to assess risk and allocate credit efficiently. This study aims to predict rural bank credit across all provinces of Indonesia, employing a clustering approach to address the diverse economic landscapes of each province. Predictive modeling will be carried out using both simple machine learning methods such as linear regression and more complex algorithms like XGBoost and LightGBM. Furthermore, these models will be stacked against each other to compare their performance and enhance predictive accuracy. The evaluation of these models will be rigorously conducted using Mean Absolute Error (MAE), Root Mean Square Error (RMSE), and Mean Absolute Percentage Error (MAPE) to ensure the reliability and precision of the predictions. The findings indicate that stacking LightGBM with XGBoost, using Linear Regression as the meta-learner, generally produces superior results across most clusters. Nevertheless, there are instances where individual models, including simpler ones like Linear Regression, excel, demonstrating their effectiveness in specific scenarios.