Monte Carlo-neural networks method for municipal solid waste forecasting: a case study of Shiraz city

This study presents a framework for municipal solid waste (MSW) forecasting using artificial neural networks (ANNs) and Monte Carlo (MC) simulations. The model integrates population, consumer price index, and temporal patterns, using data collected from the case study of Shiraz. In this approach, the model architecture parameters were treated as variables, and the randomised ANN structure was sampled via 20,000 MC draws. The optimum model, one hidden layer with 17 neurons, an 11-month input memory, and a 2-month forecast horizon, achieved strong predictive accuracy (RMSE = 1.2 × 106 kg; MAE = 1.008 × 106 kg; MAPE = 3.10%; R = 0.81). After examining the short- and long-term prediction capabilities, the model was applied in a probabilistic scenario analysis. The probability of system failure was estimated to be approximately 4.5%, indicating a non-negligible exceedance risk that warrants preventive planning and capacity-management measures. Month-wise probability density functions further showed that exceedance risk varies across the forecast horizon, highlighting the value of probabilistic forecasting for MSW capacity planning.