<p>Accurate landslide susceptibility mapping is very important in mountainous areas to reduce hazards. This study investigates how predictive modelling of landslide susceptibility prediction is enhanced by incorporating soil moisture in Minxian County, Gansu, China. To the best of our knowledge, this study describes, for the first time the usefulness of soil moisture for earthquake-induced or coseismic LSM in the study area. This study considered fourteen influencing factors associated with landslide occurrence, and previous landslide locations were split into 70% training and 30% validations datasets randomly. Three models, Random Forest RF, Convolutional Neural Network (CNN), and Logistic Regression (LR), were trained both with and without soil moisture data to confirm its contribution. The Successive Rate Curve (SRC) and Prediction Rate Curve (PRC) were used to analyze model performance, in addition to ROC-based evaluation, enabeling more accurate prediction. These metrics contribute to a more robust and dependable assessment of model performance. The results demonstrate that incorporating soil moisture improves the model’s accuracy and susceptibility outcomes. The model incorporating soil moisture performed better, and the spatial accumulation of susceptible zones was more reliable. With CNN demonstrating the highest Area Under Curve (AUC) gain (0.85 to 0.92) and RF attaining the best accuracy overall with (AUC 0.94 to 0.97), the performance gains in CNN and RF show that incorporating soil moisture and comparing multiple models directly reduced prediction uncertainty. Overall, this research has demonstrated the critical importance of soil moisture as a major regulator of earthquake-induced landslides as well as in enhancing data-driven susceptibility mapping and regional hazard management.</p>