Machine learning models have emerged as a promising approach for predicting retaining walls’ behavior, particularly when traditional analytical methods are inadequate. The estimation of earth pressure and the failure mechanisms for narrow backfill retaining walls cannot be made by the conventional earth pressure theories of Rankine and Coulomb, as they assume that the backfill is sufficiently wide enough for a failure surface to develop entirely to the ground surface. The present study uses three failure mechanisms involving one, two, and three rigid blocks depending upon the distance between the rock face and the retaining wall to generate the dataset on active earth pressures. However, these calculations are intricate and time-consuming. Therefore, the present study proposes AI/ML techniques to predict the earth pressures and failure mechanisms in narrow backfill retaining walls based on the dataset generated using analytical methods. Nine input parameters are used to train various ML models and choose a best-fit model based on various error metrics. SVR has an R2 value of 0.974 for the regression problem, and RF has an accuracy metric of 96.85% for the classification problem on the generated dataset. Interpretability of the best-fit model is presented using SHAP analysis to serve the wider geotechnical engineering audience. Machine learning models have the potential to revolutionize the design and analysis of retaining walls supporting narrow backfills.