Machine learning approaches have gained increasing attention in recent years for predicting pipe failures; however, most existing studies have relied on one or two specific models, lacking a comprehensive comparative assessment of diverse algorithms and their practical applicability.This study develops and systematically evaluates six machine learning models, logistic regression, decision tree, random forest, support vector machine, neural network, and gradient boosted tree, for pipe failure prediction. The models were constructed using the dataset from the water distribution network of District 2 in Tehran, which includes detailed physical and operational pipe attributes. A 70–30 training–validation split was employed to ensure robust model development and assessment. The results reveal that, contrary to initial expectations, more sophisticated models such as random forest and gradient boosted tree exhibited substantial overfitting, achieving near perfect accuracy on training data but demonstrating poor generalizability to validation data, thereby limiting their suitability for operational deployment. The Decision Tree model was identified as the superior option, achieving an Area Under the Curve of 0.92, a sensitivity of 93%, and an accuracy of 92%. Given its high interpretability, this model can be effectively utilized in the operational management systems of water distribution networks for predicting pipe failures.