Filter Method Feature Selection Techniques for Solid Waste Prediction Based on GRU Deep Learning Model

Municipal solid waste management (MSW) is an imperative aspect towards the development of a sustainable and healthy communities. Hence, it's important to develop a strong system to handle and dispose the MSW. In this regard, the early prediction of waste generation can play an important role in facilitating the municipal authorities for the development of a proper MSW management system and resource allocation for the collection and disposal of the MSW. Many researchers have contributed in early predictions of waste generation. However, most of the studies have chosen random variables for making predictions without considering any feature selection technique which eventually leads to over-fitting or under-fitting issue that reduces the accuracy of the model's prediction. Moreover, some researchers have conducted studies with feature selection techniques to enhance the model predictions, however, they have not focused on the comparison evaluation of different feature selection techniques to present the best technique for the MSW generation forecast. Therefore, this study has been conducted to reveal the importance of feature selection techniques, making the right choices about its selection and its influence on a model's prediction accuracy. This study has compared three most commonly used filter based feature selection techniques i.e. Pearson's correlation coefficient, mutual information, and analysis of for MSW generation forecast based on Gated Recurrent Unit model. Three distinct error matrices, namely Root Mean Square Error, Mean Absolute Error, and Regression values have been used to assess the GRU model's performance. Notably, the PCC-GRU approach demonstrated enhanced performance compared to the MIGRU and ANOVA-GRU approaches, as evidenced by achieving the lowest values for RMSE and MAE, explicitly 0.0848 and 0.0647, respectively.