Hybrid AdaBoost and CatBoost Models Enhanced with HGS and CGO for Non-Stationary Runoff Forecasting

Abstract

This study presents a hybrid forecasting model for non-stationary daily runoff series, integrating state-of-the-art machine learning algorithms, CatBoost, AdaBoost, and HGBoost, with metaheuristic optimization techniques, HGS and CGO. The model was evaluated using a dataset consisting of 29085 from Russian River Basin dataset. Performance was assessed using key metrics such as RMSE (6.79), R² (0.9549), and error scatter. Results show that the hybrid AdaBoost-HGS model outperforms other approaches, achieving the highest R² of 0.9549 and the lowest error, followed by the AdaBoost-CGO model. Although the CatBoost-HGS model showed promise, it was less effective than the other schemes. Furthermore, the study highlights the faster convergence and efficiency of AdaBoost-based models in terms of both prediction performance and computational time. These findings underscore the potential of hybrid ML models in enhancing runoff prediction accuracy and efficiency, which is crucial for applications in flood control and water resource management.