Methane Yield Prediction for Anaerobic Digestion using Machine Learning Models

Anaerobic digestion (AD) systems exhibit complex, nonlinear interactions between operational parameters, making methane yield prediction challenging using traditional analytical methods. This study evaluates multiple machine learning (ML) techniques to model methane production from an industrial reactor dataset containing operational and environmental variables. A structured workflow incorporating correlation analysis, dimensionality reduction, clustering, and supervised learning was implemented to identify key predictors and assess model performance. Reactor temperature emerged as the dominant factor influencing methane yield, while most other variables showed weak direct correlations. Principal component analysis and K-means clustering revealed distinct operational regimes associated with different performance levels. Baseline linear regression models achieved moderate predictive accuracy (R² ≈ 0.5), while nonlinear models—including ensemble methods and artificial neural networks—provided only marginal improvements, suggesting dataset limitations rather than algorithmic constraints. The results highlight the importance of richer biochemical and temporal data for improving predictive modelling and supporting data-driven optimization of industrial AD processes.