Development of a Hybrid Voting Model with SMOTE and Random Search for Classification of Religious Facility Grant Recipients

Abstract

The process of determining recipients of religious facility grants requires high accuracy to ensure that aid is distributed fairly and supports equitable community services. Manual selection methods often face challenges such as data imbalance, diverse assessment criteria, and subjective decision-making, which can reduce accuracy and efficiency. This study proposes a hybrid machine learning model using Voting Ensemble (Hard and Soft), combining Logistic Regression (LR), Support Vector Machine (SVM), and K-Nearest Neighbor (KNN), optimized with Random Search and supported by SMOTE to handle class imbalance. The dataset consists of religious facility grant applications in Riau Province, with preprocessing, SMOTE balancing, and Stratified K-Fold Cross Validation applied for robust evaluation. The experimental results show that the Hybrid Voting model outperforms single models, achieving an average accuracy of 99.46%, with precision, recall, and F1-score consistently above 96%, and some folds achieving 100% accuracy. These findings demonstrate that the hybrid approach enhances prediction stability, reduces misclassification of minority classes, and provides a decision-support system that is objective, accurate, and efficient for grant recipient selection.