Analisis Penerapan Algoritma Naive Bayes dengan Seleksi Fitur untuk Klasifikasi Penerima Program Sembako di Kabupaten Banyuwangi

Abstract

The inaccuracy of targeting in the distribution of the Sembako Program remains a challenge in Banyuwangi Regency. This study aims to analyze the application of the Naive Bayes algorithm with Information Gain feature selection for classifying Sembako Program recipients and to evaluate the effect of feature selection on classification performance. The study used secondary data from the March 2025 National Socio-Economic Survey (Susenas) in Banyuwangi Regency, consisting of household, individual, and consumption modules. After data preprocessing, 949 household records with 61 features were obtained. The target variable was the status of Sembako Program recipients based on the Pernah_Menerima_BPNT variable. To address class imbalance, Random Under-Sampling was applied using 1:1 and 1:2 sampling ratios. Feature selection was performed using Information Gain with five feature subsets, namely Top 5, Top 8, Top 10, Top 12, and Top 15, followed by classification using the Multinomial Naive Bayes algorithm. Model performance was evaluated using accuracy, precision, recall, F1-score, confusion matrix, and ROC-AUC. The experimental results show that the highest accuracy of 67.52% was achieved using the Top 5 features with a 1:2 sampling ratio. However, the best overall model based on F1-score was obtained using the Top 10 features with a 1:1 sampling ratio, achieving an AUC value of 0.65, indicating a fair classification performance in distinguishing Sembako recipients from non-recipients. These findings demonstrate that Information Gain effectively improves the performance of the Naive Bayes classifier, while the selection of an appropriate sampling ratio influences the balance between overall accuracy and the detection of minority-class instances.