Data-driven optimization of rice husk waste management through an integrated machine learning and community-based pyrolysis approach

Background: Indonesia’s energy landscape currently pivots between two bifaceted issues: the stagnation of the national energy transition and the inefficiencies of decentralized waste management. Despite East Java producing 9.27 million tons of dry-milled rice (GKG) in 2024, the resulting 1.85 Mt of rice husk remains an underutilized bio-resource. This wasted potential coincides with a sluggish renewable energy trajectory, where the 15.25% share by mid-2025 significantly trails the 23% national target. Methods: A data-driven framework integrating feedstock characterization, experimental data, and literature benchmarks was applied to evaluate catalytic fast pyrolysis and upgrading pathways for rice husk. Machine-learning-assisted correlation analysis and multi-objective optimization (NSGA-II) were used to benchmark key process variables, product yields, and fuel quality trade-offs. Findings: The technical foundation, built on detailed feedstock characterization, reveals that the CFP process yields ~46.9 wt% bio-oil, which is further refined to a 32.2 wt% biodiesel-equivalent yield. To enhance operational precision, various ML algorithms were evaluated; the Extra Trees model coupled with Non-dominated Sorting Genetic Algorithm II (NSGA-II) demonstrated superior predictive performance with an R2 of up to 0.96 and an RMSE