Machine learning-driven prediction of nitrogen loss in organic solid waste composting

Composting represents a crucial component of sustainable waste management, providing significant resource recovery and environmental advantages. However, nitrogen loss during composting remains a significant challenge, necessitating the development of a predictive model for nitrogen loss during the composting process. This investigation implemented five machine learning models, utilizing 307 data points encompassing composting strategies, physicochemical properties, and composting time stages, to predict nitrogen loss during organic solid waste composting. The findings demonstrated that the adaptive boosting (AdaBoost) algorithm achieved optimal performance with a coefficient of determination of 0.847 after eliminating redundant features (scale and C/N). Moreover, Shapley additive explanation analysis identified several key factors significantly influencing nitrogen losses during composting, including composting time stages, bulking agents, raw materials, and ammonium nitrogen levels. Notably, the initial phase of composting emerged as the most critical period for nitrogen loss. The utilization of sawdust, rice husk, and corn stalk as bulking agents enhanced nitrogen retention in compost. Furthermore, implementing static aeration for ventilation and applying chemical additives effectively reduced nitrogen losses during the composting process. These results provide a scientific foundation for identifying optimal composting conditions to minimize nitrogen loss, thereby offering practical guidance for effective composting operations.