Automated Decision Support System for Social Assistance Eligibility Based on House Images Using Deep Learning

Abstract

Social assistance is one of the government programs aimed at improving the welfare of underprivileged communities. However, the process of determining eligible beneficiaries is still largely conducted manually, leading to subjectivity, inaccurate targeting, and time-consuming decision-making. This study proposes an automated decision support system for determining social assistance eligibility based on house images using Deep Learning and the Analytical Hierarchy Process (AHP). The proposed system consists of a house image classification model developed using the MobileNetV2 architecture, a web-based application developed with the Laravel framework, and the integration of image classification results with the AHP method. MobileNetV2 is employed to classify house conditions into eligible and ineligible categories while generating confidence scores. These confidence scores are converted into a 1–10 assessment scale and used as the House Condition criterion in the AHP calculation together with income, occupation, number of dependents, and house ownership. The study utilized a dataset of 300 house images for model training and evaluation. Experimental results show that the MobileNetV2 model achieved an accuracy of 74.00%. Furthermore, the developed system successfully integrates automatic house image classification with AHP-based decision-making, producing more objective, consistent, and accurate recommendations for social assistance recipients while assisting local governments in improving the efficiency and transparency of the beneficiary selection process.