Oversampling-based prediction of environmental complaints related to construction projects with imbalanced empirical-data learning

In this study, an oversampling-based method for optimally predicting complaints due to environmental pollutants from construction sites was developed using imbalanced empirical data. Based on the empirical database related to such environmental complaints, several environmental-complaint prediction models were generated. These took into consideration input data, oversampling, and machine learning. An optimal prediction model with the best performance was determined using an integrated multi-objective optimization algorithm. To confirm the feasibility of the developed oversampling-based method, the optimal prediction model was determined among 210 other generated prediction models in the case study. By transforming imbalanced data on such environmental complaints into balanced data through oversampling, the prediction models overcame limitations (biased results) of imbalanced-data prediction, and recorded an 8–23% performance improvement. The optimal prediction model predicted imbalanced environmental complaints more efficiently and accurately than the previous other prediction models: up to 66.5% performance improvement. Ultimately, the optimal prediction model is expected to serve as a starting point for an environmental-complaint management system creating sustainable communities by minimizing the damage caused by environmental complaints to people living near construction sites and concerned construction companies.