Classification Of Social Assistance Eligibility In Talang Benuang Village Using C4.5 And Naive Bayes

Abstract

The selection of Non-Cash Food Assistance (BPNT) recipients in Talang Benuang Village is still performed manually, creating a risk of mistargeted distribution. This study develops and compares C4.5 and Naive Bayes models for classifying BPNT eligibility. A data-driven approach followed CRISP-DM, using 100 labeled public records for modeling and 50 village household records for application. The analysis comprised data cleaning, ordinal transformation, pipeline construction, C4.5 hyperparameter tuning with GridSearchCV, and 5-fold cross-validation evaluated through a confusion matrix and ROC-AUC. Both algorithms achieved 93.00% accuracy, 100.00% precision, 85.42% recall, and a 92.13% F1-score. C4.5 was selected as the final model because it produced interpretable decision rules and an AUC of 0.986, exceeding the Naive Bayes AUC of 0.868. Application to 50 households yielded 42 eligible and 8 ineligible predictions. These findings indicate that C4.5 can support objective preliminary screening of BPNT recipients; however, administrative decisions still require verification through village deliberation and official ground-truth records.