<![CDATA[Inaccurate targeting in subsidized LPG distribution remains a persistent policy challenge in Indonesia, where manual verification processes are vulnerable to misuse and administrative error. Addressing this gap, the present study develops and evaluates a biometric identity verification system based on Convolutional Neural Networks (CNNs) to improve the accuracy and accountability of subsidy allocation at the point of distribution. Following the CRISP-DM framework, two CNN architectures with fundamentally different design philosophies were compared: ResNet-IR, optimized for representational depth and recognition accuracy, and MobileFaceNet, designed for computational efficiency on resource-constrained hardware. Both models were sourced from the InsightFace framework as pre-trained models and evaluated on a locally acquired dataset of 111 registered subsidy recipients from Pajang Village, Tangerang City. Evaluation across face identification (1:N) and face verification (1:1) tasks reveals that ResNet-IR consistently outperforms MobileFaceNet, achieving an accuracy of 94.7% with a precision, recall, and F1-score of 0.9043, compared to MobileFaceNet’s accuracy of 93.7% and F1-score of 0.8862. The primary contribution of this work is to demonstrate, for the first time in the Indonesian subsidy distribution context, that deep learning-based facial recognition can serve as a viable, deployable mechanism for biometric identity verification in public service programs offering a technically grounded pathway toward more transparent and equitable subsidy targeting.]]>
