IMPLEMENTASI SISTEM KLASIFIKASI PENDUDUK PEMERINTAH DAERAH MENGGUNAKAN ALGORITMA ARTIFICIAL NEURAL NETWORK (ANN)

Abstract

This study aims to classify residents based on 29 parameters with two labels, namely eligible (label 1) and not eligible to receive assistance from the local government (label 0). Tests were conducted using imbalance data handling methods, namely undersampling and oversampling with epochs variations of 100, 200 and 300. Data trained using neural network algorithms in the default state, resulted in more than 95% accuracy in both train and test data. However, it showed low results on precision, recall and f1-score metrics for label 1. Subsequent tests with oversampling and undersampling produced almost as good accuracy as the previous tests but showed significant improvement in the precision and recall metrics, with a value of 96% on label 1 in both metrics when oversampling tests. The increase in precision and recall metrics also occurred in the undersampling test, where the precision and recall metrics obtained were 76 and 84% at label 1. However, overall, the oversampling method with 100 epochs showed the most optimal performance but without experiencing extreme overfitting on data with label 1.