Trustworthy and Human Centric neural network approaches for prediction of landfill methane emission and sustainable waste management practices.

Landfills rank third among the anthropogenic sources of methane gas in the atmosphere, hence there is a need for greater emphasis on the quantification of landfill methane emission for mitigating environmental degradation. However, the estimation and prediction of landfill methane emission is a challenge as the modeling complexity of methane generation involves different chemical, biological and physical reactions. Various machine learning techniques lacks in providing explainability and the context in addressing uncertainties of landfill emission. This work presents novel artificial neural network (ANN) based approach for enhancing the interpretation of methane prediction. A trustworthy ANN (TANN) using SHapley Additive exPlanations (SHAP) is presented in this research work improving the explainability of predicted values of methane emission using the emission data of seven major methane producing countries like India, China, Russia, Indonesia, US, EU, Brazil. Further, a Human-Centric ANN (HCANN) model based on two approaches: environmental methane emission risks indication and physics informed model are developed. The HCANN was capable of learning scientific principles and addressing modeling complexity using the well-known LandGEM emission model data. The prediction results of physics informed model exhibited close agreement with those produced by LandGEM. Likewise HCANN model developed using the factors like methane production rates (MPR), gas capture system efficiency (GCSE), monitoring system reliability (MSR) is able to offer intuitive and contextual decision and to understand the risk associated with unmanaged methane. Proposed TANN and HCANN approaches offer valuable tool for prediction of methane emission and risk assessment for sustainable waste management practices.