Abstract: Objective Regional landslide susceptibility assessment serves as a critical scientific basis for formulating disaster prevention and mitigation policies, playing a key role in safeguarding lives and property. Traditional evaluation systems often struggle to effectively quantify the impact of extreme precipitation on landslides. This study, taking Yulin City, Shaanxi Province as the study area, aims to develop novel coupled models that integrate extreme precipitation indices to enhance the accuracy and practicality of landslide susceptibility assessment. Methods We innovatively introduced Extreme Precipitation Indices (EPIs) and coupled them with three machine learning algorithms—Random Forest (RF), Categorical Boosting (CatBoost), and Support Vector Machine (SVM)—using the Information Value (IV) model. This led to the construction of three integrated models: EPIs-IV-RF, EPIs-IV-CatBoost, and EPIs-IV-SVM, through which a systematic landslide susceptibility assessment was conducted. Results The EPIs-IV-CatBoost model yielded more rational spatial zonation, with very high and high susceptibility areas concentrated in the loess hilly region of southeastern Yulin, accounting for 4.58% and 14.33% of the total area, respectively;The incorporation of EPIs significantly improved model accuracy, with the EPIs-IV-CatBoost model performing the best (AUC = 0.937); SHAP analysis revealed that extreme precipitation indices and slope are among the most important influencing factors, playing a critical role in landslide development in the region. Conclusions The EPIs-IV-CatBoost coupled model developed in this study demonstrates reasonable zonation and high accuracy, providing a reliable quantitative tool and scientific support for regional disaster prevention and mitigation planning under extreme precipitation scenarios.