Abstract: The Gradient Boosting Model (GBM) has demonstrated strong predictive performance in landslide susceptibility assessment; however, the complexity of factors limits its stability and accuracy. To improve the model's accuracy in susceptibility assessments from the input level, this study explored the optimization effect of the Certainty Factor (CF) coupling method on the GBM. using Yongjia County in Zhejiang Province as a case study, 10 evaluation factors were selected based on Spearman correlation coefficient and Shapley value. The CF coupling method was applied to optimize four GBM models: Ada, Gentle, Logit, and RUS, which were then compared. The results show that the factors most significantly affecting landslides in the study area, in descending order, are: elevation, normalized vegetation index, roads, slope, water systems, land type, sediment transport index, vegetation type, lithology, and rainfall. Optimization using the CF coupling method improved the Kappa coefficient of GBM models by 0.04 to 0.14. ROC (Receiver Operating Characteristic) curve validation showed that, except for the adaptive boosting model with a smaller performance increase, the AUC (Area Under Curve) values of the other models increased by 1% to 7%, with the CF-Logit Boosting model achieving the highest accuracy. The optimized models demonstrated better distinction between very low and very high susceptibility areas, indicating that the CF coupling method can enhance both the accuracy and stability of the models. The GBM models optimized by CF coupling method could provides a theoretical basis for reducing data complexity and improving the accuracy of landslide susceptibility assessments.