Multivariate Time Series Forecasting of Municipal Solid Waste

Municipal solid waste is a major concern these days. With industrialization and urbanization, the amount of solid waste being generated has significantly increased over the years. This waste is mostly disposed to landfills and only a proportion of it is recycled. The waste on landfills has detrimental effects on health. Hence, it is important for the government to take appropriate measures by having enough recycling sites for regular treatment of solid waste. However, setting up of recycling units requires investment. Therefore, it is important to forecast waste that will be generated in coming years and at the same time predict the amount of waste that will be recycled. If the current recycling units are inadequate to handle the future quantities of waste generation, then probably more such units can be set up. Thus, this requires better forecasting of solid waste that will be generated and recycled. This study uses time series data to forecast waste generation and recycling. The study uses statistical models ARIMAX and machine learning models such as random forest, support vector regression, XGBoost for forecasting. The study also builds hybrid models by combining these models to find a model that provides higher accuracy, low error rate and thus better prediction.