The calculation of the non-linear correlation between the landslide inventories and their environmental factors is an important factor that affects the uncertainty of the landslide susceptibility prediction (LSP) modeling. In order to study the changing patterns of LSP under the influence of the uncertain factors, taking Yanchang County of China as example, 82 landslides and 14 environmental factors are obtained, and the frequency ratio (FR) and weight of evidence (WOE) connection methods are coupled with the chi-squared automatic interaction detector (CHAID) decision tree model to carry out LSP. Then the original environmental factors data (hereinafter referred to as "original data") is used as the input variable to compare the individual CHAID decision tree model to realize the analysis of LSP modeling pattern. ROC accuracy, mean, standard deviation, and average rank are adopted to analyze the uncertainty characteristics in the LSP modeling process. Results show that:1) LSP uncertainty of the WOE-CHAID model is lower than that of the FR-CHAID model, and WOE has relatively excellent nonlinear correlation performance. 2) The prediction accuracy of individual CHAID decision tree model is slightly lower than that of the WOE-CHAID and FR-CHAID models, but it has higher modeling efficiency. 3) In terms of reflecting the spatial correlation between landslides and its environmental factors, the CHAID decision tree model coupled with FR and WOE connection methods have significant advantages. Generally, WOE is a better connection method and CHAID decision tree model has good prediction performance and high prediction efficiency. Susceptibility prediction by the WOE-CHAID decision tree model is less uncertain and more in line with the actual landslide probability distribution characteristics.