Perbandingan Convolutional Neural Network dan Algoritma Machine Learning Konvensional untuk Klasifikasi Kemiskinan Multidimensional di Indonesia

Abstract

Multidimensional poverty in Indonesia is a complex phenomenon involving various interconnected social, economic, and structural aspects. Conventional approaches to poverty classification often fail to capture non-linear interaction patterns and spatial dependencies inherent in multidimensional socio-economic data. This research aims to compare the performance of Convolutional Neural Networks (CNN) with conventional machine learning algorithms such as Random Forest and XGBoost in classifying multidimensional poverty in Indonesia. The research method employs a comparative quantitative approach using data from the 2023 National Socio-Economic Survey (Susenas) by BPS, covering 8,000 household observations. The target variable is multidimensional poverty status based on the Multidimensional Poverty Index (MPI) with a 1/3 cutoff. Data was split 70:30 for training and testing, with preprocessing including missing value imputation, one-hot encoding, and Min-Max scaler normalization. The CNN model was designed with a two-convolutional-layer architecture, while Random Forest used 200 decision trees and XGBoost with 200 estimators. Research results demonstrate that CNN provides the best performance with 82.4% accuracy, outperforming Random Forest (80.1%) and XGBoost (81.2%). Important variable analysis reveals that housing infrastructure conditions, household head education level, and sanitation access are key factors in determining multidimensional poverty, providing strategic input for formulating more targeted poverty alleviation policies.