Predicting chemical composition of corn silage using tree-based machine learning models: Model development, feature analysis, and practical application.

This study investigated the associations between corn raw material/process parameters and chemical components after anaerobic fermentation using machine learning (ML). High-quality experimental data from 140 academic papers were utilized for model training, with 19 input variables-including corn raw material components and process parameters. Eleven high-accuracy models were successfully constructed to predict the nutritional components and fermentation quality of corn silage after fermentation, achieving a maximum R2 value of 0.98. The interpretability of the tree-based models provide new insights into the anaerobic fermentation process in corn. Based on the optimal model, a user-friendly and interactive prediction platform (http://silagedb.com/SMART-Maize/) is freely available to facilitate scientific research. In summary, this study utilizes ML to demonstrate that raw materials and processing parameters determine the post-fermentation quality. Furthermore, the proposed methodological framework can inform future model for other anaerobic fermentation processes.