The increase in water consumption and demand has highlighted the need to improve the efficiency of water distribution networks (WDNs). Pumping stations (PS) represent a significant challenge due to their high energy consumption and associated operational costs. This study presents a data-driven evolutionary optimization methodology focused on predicting cost and penalties in the design and operation of PS. The methodology integrates the functioning of genetic algorithms with machine learning techniques, developing a surrogate model capable of predicting associated costs and increasing the computational efficiency of optimization models. A case study is presented to validate the methodology, showing a significant reduction of 31.8% in evaluation times and a decrease in optimization time compared to traditional models. The results indicated that despite the reduction in computational effort, the increase in the final cost of the network was minimal, with only a 2.42% increase compared to the baseline model. These findings underscore the effectiveness of combining machine learning with genetic algorithms for the optimization of PS in WDNs, improving computational efficiency while maintaining high standards in solution quality.