Design optimization for widely used axial flow pumps presents a formidable challenge due to the significant impact of numerous parameters associated with impeller geometry on hydraulic performance. The expansive design space raises concerns about the cost and time implications of the optimization process. This paper introduces a machine learning-based algorithm with a dynamic sampling approach to enhance the hydraulic performance of axial flow pumps. The focus is on an axial flow pump designed for China’s South-to-North Water Diversion Project. Optimization involves selecting 15 design variables governing impeller geometry, considering meridional shape and mean blade profiles. The optimization process predicts hydraulic performance using CFD methods, with a primary objective of maximizing efficiency at the axial flow pump’s design point while maintaining pump head around the design value. The results indicate that the proposed machine learning-based algorithm exhibits commendable convergence, delivering a notable improvement in performance. For instance, the optimized axial flow pump displays 2% efficiency increase compared to the initial design. Further analysis employing concepts like entropy generation rate and boundary vorticity flux reveals that the optimized pump has more uniform flow near the pressure side of the impeller blade. Additionally, design optimization effectively suppresses flow separation at the blade trailing edge near the impeller hub. This study offers valuable insights and a practical tool for the design optimization of axial flow pumps.