Machine Learning Approaches for Municipal Solid Waste Generation Forecasting

Abstract

Municipal solid waste (MSW) generation forecasting can be considered as the biggest challenge of integrated solid waste management systems, particularly for developing countries where data collection is limited. In this study, three different machine learning algorithms, namely backpropagation neural network (BPNN), support vector regression (SVR), and general regression neural network, were applied for different countries. Comparative evaluation of these different algorithms based on gross domestic product, domestic material consumption, and resource productivity were given through the optimum solution. Moreover, the algorithms were tested for the case of Turkey. The results of this study are expected to represent a general outline for stakeholders of Turkey for improving MSW management strategies all over the country, and these results can be extended to similar developing countries across the world. It can be concluded that BPNN and SVR methods can be applied successfully for the case of Turkey and other countries across the world to predict the MSW generation, whereas BPNN is slightly better. If the input and output variables are identified well, machine learning approaches can give a good projection for waste generation, and this projection can be utilized for different countries. Furthermore, the developing countries with missing data can develop more realistic strategies for MSW management by not relying solely on international databases such as Eurostat to forecast MSW generation.