Journal of Engineering Research and Reports

The improper management of hazardous waste poses severe risks to human health and the environment, making effective regulatory enforcement a critical priority for environmental agencies. The Resource Conservation and Recovery Act (RCRA) governs hazardous waste management across the United States, yet traditional inspection-targeting methods—based on cyclical schedules or facility size alone—are insufficient to proactively identify non-compliant generators. This study presents a data-drivenf predictive compliance modeling framework that integrates multi-source regulatory datasets, including EPA RCRA Info, Biennial Reports, Annual Reports, and Tier Two chemical inventory data, to predict which hazardous waste generators are most likely to be out of compliance in the next reporting cycle. Two ensemble machine learning algorithms, Random Forest (RF) and Gradient Boosting (GB), were evaluated on six engineered feature categories: past violation history, total annual waste quantity, number of waste streams, facility size, time since last inspection, and waste toxicity. Feature importance analysis identified past violation history as the dominant predictor (importance = 0.35), followed by total annual waste quantity (0.25) and number of waste streams (0.18), reflecting recidivism and operational complexity as the primary drivers of non-compliance. Performance simulations grounded in published EPA and NYSDEC field test benchmarks indicate that the proposed framework can improve inspection hit rates by 46–79% over conventional targeting strategies. The framework is portable across jurisdictions and provides a scalable, interpretable, and auditable tool for modernizing risk-based RCRA enforcement.