<p>The settlement prediction around foundation pit is crucial for ensuring the safety and stability of urban construction projects. However, existing studies often face challenges such as limited interpretability of machine learning (ML) models and the inability to perform multi-target predictions for complex geotechnical engineering schemes. To address these issues, this paper investigates the application of explainable machine learning (EML) techniques for the settlement prediction around foundation pit, using measured data from the Yongning Hospital project in Huangyan District as a case study. Two predictive schemes, time series prediction and random sampling prediction, were proposed and validated using classical machine learning models, including multilayer perceptron (MLP), random forest (RF), and extreme gradient boosting (XGBoost). The results indicate that the random sampling prediction scheme, with the RF model as the dominant predictor, achieves high accuracy, with an average root mean square error (RMSE) of 0.110 (mm) and a coefficient of determination (<i>R</i>^<i>2</i>) of 0.985. Feature importance ranking and SHapley Additive exPlanations (SHAP) analysis reveal that critical features, such as the top displacement of the enclosing pile, significantly influence settlement predictions, enhancing the model’s interpretability. Furthermore, optimization of input features and output targets reduced the number of input features by 70%, lowering resource expenditure while maintaining acceptable accuracy. This research advances geotechnical engineering practices by promoting the use of EML to enhance the accuracy, transparency, and efficiency of the settlement prediction around foundation pit.</p>