Machine learning analysis of biogas production using waste as a resource

This study aims to use machine learning techniques to examine the production of biogas from waste materials. Our goal is to forecast important output characteristics including biogas yield, biogas generated, CH4 yield, and biodegradability using experimental data on biogas production. Waste type, pre-treatment, co-digestion status, reactor type, and operational parameters (such as TS/VS ratio, C/N ratio, temperature, pH, and hydraulic retention length) are among the input features that are taken into account in this study. The dataset includes experimental data gathered from several studies that cover a range of waste types, including wastewater, food waste, agricultural waste, municipal solid waste (MSW), and organic fraction of municipal solid waste (OFMSW). These inputs are utilized for training machine learning models, such as Random Forest (RF), to predict output variables. Prior to model training, the dataset was cleaned and organized using preprocessing techniques such feature selection, and missing value treatment. Boruta analysis was also performed to decide on input parameters. High predictive performance was demonstrated by the data, especially for biogas generated (R2 = 0.92) and yield (R2 = 0.94). Operational parameters, particularly working volume and temperature, have a considerable impact on production projections, according to feature importance analysis. This study provides important insights for upcoming waste-to-energy systems and shows how machine learning may be used to improve biogas production processes. In order to increase model precision, future research can concentrate on adding more input and output parameters. It can also investigate the use of additional machine learning approaches to improve performance.