Enhancing Runoff Prediction through Feature Engineering and Cluster-Specific Modeling

Abstract

Accurate runoff prediction plays a crucial role in water resource management, flood control, and hydropower generation. This study proposes a novel hybrid regression approach for runoff prediction by integrating clustering techniques as a preprocessing phase with regression algorithms. Initially, the dataset is divided into distinct clusters to capture underlying patterns in the runoff data. Subsequently, a regressor model is trained on each cluster to enhance predictive performance. The proposed methodology is evaluated using real-world hydrological datasets, and its effectiveness is compared against baseline models. Experimental results demonstrate that clustering-based modeling improves prediction quality, as indicated by key performance metrics such as RMSE and R2. The findings suggest that the hybrid regressor method can significantly enhance the reliability of runoff predictions, offering valuable insights for hydrological forecasting and water management applications.