Cosgun, AhmetGunay, M. ErdemYildirim, Ramazan2026-04-042026-04-0420250360-31991879-3487https://doi.org/10.1016/j.ijhydene.2025.04.213https://hdl.handle.net/11411/10450In this work, tri-reforming (TRM) of biogas was investigated using a variety of machine learning (ML) tools for knowledge extraction. For this purpose, a comprehensive database including 1183 data entries with 41 descriptors and 3 performance measures (CH4 conversion, CO2 conversion, and H2/CO ratio) was compiled from 29 articles published between 2004 and 2024. Random forest (RF) models were constructed to predict the values of performance measures that can be obtained under unknown conditions; the models were usually quite successful in the majority of the cases with the training/testing R2 values of 0.99/0.87, 0.99/0.91 and 0.96/0.58 for CH4 conversion, CO2 conversion, and H2/CO ratio respectively. To bring some explainability to the predictive models, the SHapley Additive exPlanations (SHAP) analysis was performed to determine the importance of descriptors and their effects on the performance measures. Among many results, SHAP analysis of CH4 conversion revealed the most important variable to be the reaction temperature, followed by calcination time, H2O and O2 percentages in the reaction stream, and W/F ratio. Lastly, to improve the explainability of ML even more, DT classification analysis was successfully used to generate heuristic rules that describe the combinations of individual descriptors leading to different levels of the target variables.eninfo:eu-repo/semantics/closedAccessBiogas UtilizationTri-ReformingMachine LearningRandom Forest RegressionDecision Tree ClassificationArticle2-s2.0-10500249110510.1016/j.ijhydene.2025.04.21310.1016/j.ijhydene.2025.04.213607Q1595127Q1WOS:001472214500001