Forecasting residential and nonresidential solid waste generation, disposal, and diversion using three machine learning approaches

Abstract Accurate forecasting of solid waste quantities is essential for sustainable waste management planning, yet limited research exists in this area. This study develops a framework to forecast solid waste generation, disposal, and diversion quantities using three machine learning (ML) approaches: artificial neural networks (ANNs), support vector machines (SVMs), and multiple linear regression (MLR) models. The forecasting framework is based on 12 socioeconomic variables, the values of which were derived from publicly available data sources. Projections for 2023 to 2050 were developed considering data preprocessing, training, and testing, to create reliable datasets. Correlation analysis was used to rank predictor and response variables, and statistical tests were conducted to identify heteroscedasticity and linear relationships. A case study was conducted for Canada and four provinces: Alberta (AB), British Columbia (BC), Ontario (ON), and Quebec (QC). The results show that ML algorithms predict solid waste effectively, achieving coefficients of determination ( R 2 ) of 99.9% with ANNs and 98.6% with SVMs. The total waste generation for Canada, forecast through ANNs, SVMs, and MLRs, increased by 18.29%, 22.45%, and 22.61%, respectively, in the 28 years from 2023 to 2050. In 2050, the projected values of waste generation using the three methods were 43.67, 45.14, and 44.47 million tonnes, respectively, in Canada. ANN forecasts for 2050 project 7.75 million tonnes in AB, 5.36 in BC, 17.85 in ON, and 8.81 in QC. Waste generation is increasing with increasing population size. The method developed here can be used globally with appropriate data adjustments. The results can help in policy development and decision making.