Classification Index Village Building (IDM) Using Random Forest And Gradient Boosting

Abstract

Village development requires mapping conditions based on the Village Development Index (IDM) into categories of Very Underdeveloped, Underdeveloped, Developing, Advanced, and Independent. This study aims to develop and evaluate a classification model for IDM status in Probolinggo Regency using the Random Forest and Gradient Boosting algorithms integrated with the Synthetic Minority Over-sampling Technique (SMOTE) and GridSearchCV. SMOTE was applied to address data imbalance, while Grid Search was used for hyperparameter optimization and feature importance analysis. The results showed that Random Forest produced better classification performance than Gradient Boosting before optimization, with accuracies of 93.10% and 89.66%, respectively. After optimization using GridSearchCV, the accuracy of Gradient Boosting increased to 93.10%, while Random Forest remained at 93.10%. These results indicate that hyperparameter optimization has different effects on each algorithm. The application of SMOTE and GridSearchCV can support the IDM status classification process on data with an unbalanced class distribution. The use of IDM indicator variables as well as the Village Fund ceiling and realization provides additional information in the classification process, while feature importance is used to identify the relative contribution of variables to the model.