<p>Communication towers are critical infrastructure for mobile networks and play a crucial role in maintaining the reliability of communication systems. Currently, the condition assessment of communication towers mainly relies on periodic manual inspections, which have limitations in real-time responsiveness and inspection automation. In this study, explainable machine learning (ML) algorithms were applied on a comprehensive dataset collected by on-site inspection of 1187 communication towers in Heilongjiang Province, China. Binary labeling and a weighted-scoring method were applied to represent the damage severity of the structural components on the towers. <i>k</i>-nearest neighbors (<i>k</i>-NN), random forest (RF), artificial neural networks (ANN) and extreme gradient boosting (XGBoost) were trained to predict the overall structural condition levels of communication towers. SHapley Additive exPlanations (SHAP) analysis was applied to identify and quantify key features which have major impacts on the model predictions. The results indicate that all four models perform better on the dataset processed by the weighted-scoring method than that processed by the binary labeling. Among the four models, the XGBoost performed the best with a classification accuracy of 92.74% on the test set. From explainable analysis, “Tower column and web member” and “Verticality” were found to be the key features for towers to be classified as minor, moderate and severe failure. For condition level of extreme failure, “Tower bolt” and “Foundation” were the most critical factors. By integrating explainable ML with traditional manual inspection methods, this study provides a framework for the assessment of the structural safety of communication towers and supports maintenance decision-making.</p> Graphical abstract <p></p>