Abstarct: As the challenges surrounding water resources continue to grow, effectively managing runoff in watersheds is becoming increasingly critical for ensuring the sustainability of these resources. In this study, we focus on simulating runoff in the Kan watershed, a region that faces significant water management challenges. The performance of two hydrological models—SWAT and HBV—was analyzed over a period spanning from 1992 to 2019, with the goal of determining how well each model can simulate runoff. This time span was divided into two phases: calibration from 1992 to 2013, followed by validation from 2014 to 2019. The central question of this research was not just whether these models could accurately simulate runoff, but also whether their accuracy could be enhanced by applying advanced machine learning techniques such as the XGBoost algorithm. Results indicated that both models performed well when it came to simulating baxseflow, but they encountered difficulties in accurately modeling peak runoff events, a crucial factor in flood management. The SWAT model, overall, demonstrated stronger performance, yielding a KGE of 0.74, RMSE of 2.44, and PBias of 7.57. Meanwhile, the HBV model, though useful, exhibited somewhat lower performance metrics, with a KGE of 0.57, RMSE of 2.61, and PBias of 18.96. To address these shortcomings, hybrid models incorporating the XGBoost algorithm were developed, resulting in the SWAT.XGBoost and HBV.XGBoost models. These enhanced models demonstrated considerable improvements in performance, especially in the simulation of peak flows. Specifically, the SWAT.XGBoost model produced a KGE of 0.78, RMSE of 1.96, and PBias of 1.01. By contrast, the HBV.XGBoost model, while improved, still lagged behind with a KGE of 0.56, RMSE of 2.42, and PBias of 17.21. In conclusion, while both original models provided valuable insights, it is evident that the integration of machine learning techniques, particularly XGBoost, significantly improved accuracy. This was especially apparent in the more complex scenarios involving peak flow simulations. Therefore, the hybrid models offer a promising avenue for more reliable hydrological predictions, which could greatly aid in water resource management and flood mitigation strategies moving forward.