Implementation of IoT and Artificial Intelligence for Real-Time Monitoring of Subsidized LPG Distribution

Abstract

Subsidized liquefied petroleum gas (LPG) distribution in Indonesia faces persistent challenges in targeting accuracy, transaction transparency, and identity-based control at retail outlets. This paper proposes a novel Artificial Intelligence of Things (AIoT)-based monitoring architecture that advances beyond existing rule-based approaches by integrating a hybrid anomaly detection module combining rule-based validation with an Isolation Forest machine learning model for transaction irregularity detection. Unlike prior IoT-LPG studies focused on physical safety, our system addresses retail-level governance by validating consumer identities via NIK/KK numbers, enforcing daily purchase limits, and flagging anomalous transaction patterns. An artifact-oriented qualitative design guided prototype development, evaluated through 12 scenario-based black-box test cases covering normal purchases, repeated purchase attempts, invalid identities, stock-out scenarios, and cross-outlet duplicate detection. Results demonstarte successful execution of all predefined functional test scenarios under controlled conditions, including successful anomaly flagging and demonstrable ML model integration for streaming detection. The system remains a conceptual prototype pending integration with official population databases and field deployment across diverse outlet conditions.