Work

Data‐Centric Evaluation of Machine Learning for Biogas Production Prediction

Year: 2026

Type: article

Abstract: Reliable prediction of biogas production is important for the planning and operation of anaerobic digestion facilities, yet the predictive performance reported in the literature is often evaluated using a single hold‐out test set, providing limited insight into model generalization. This study presents a data‐centric evaluation of machine learning (ML) approaches for biogas production prediction using plant‐level information obtained from the EPA AgSTAR database and meteorological variables derived from the Meteostat platform. Four algorithms, namely linear regression, random forest, gradient boosting, and extreme gradient boosting (XGBoost), were trained and evaluated using both conventional hold‐out testing and five‐fold cross‐validation. While ensemble models achieved relatively strong hold‐out performance ( R 2 = 0.57–0.81), substantially lower cross‐validation scores ( R 2 = 0.31–0.36) revealed a considerable generalization gap. Feature‐importance analysis indicated that operational characteristics and livestock‐related variables contributed most strongly to prediction performance, whereas environmental variables showed negligible influence under the current feature‐construction strategy. Further analyses, including subgroup evaluation and bootstrap‐based uncertainty estimation, suggested that dataset size, heterogeneity, and limited environmental representativeness constrained predictive reliability. Rather than identifying limitations of ML itself, the findings highlight the importance of data quality, geographic representativeness, and robust evaluation practices in biogas production prediction. The study demonstrates that improvements in data collection and environmental‐data integration are likely to contribute more substantially to reliable generalization than additional algorithmic complexity when working with datasets of this type.

Source: International Journal of Energy Research

DOI: 10.1155/er/9261434

Author: Zeydin Pala

Institution: Muş Alparslan University

Language: English

Topic: Anaerobic Digestion and Biogas Production

Subfield: Building and Construction

Field: Engineering

Domain: Physical Sciences

SDGs: Affordable and clean energy, Climate action

Open Access status: gold

Estimated APC paid: $2,700