The West Sichuan Railway in China traverses steep terrain and is prone to moderate to strong earthquakes. The geological complexity of the railway's surroundings leads to the occurrence of geological disasters, including landslides, collapses, and mudslides during construction and operation. These landslides pose a grave risk to passenger safety and railway infrastructure. Along the Chengdu Baiyu section of the railway alone, there have been 126 reported landslides. Landslide danger encompasses the possibility of slopes transforming into various types of disasters under the influence of multiple factors. Conducting risk assessments for landslides is crucial for effective prevention and control, with the selection of suitable evaluation factors playing a pivotal role in the process. Numerous studies have explored different approaches for selecting evaluation factors and developing models for landslide risk assessment. For instance, one study employed techniques like rough set analysis, correlation analysis, and principal component analysis to screen landslide evaluation factors and utilized support vector machine models for susceptibility evaluation. This approach demonstrated that reducing evaluation factors enhances the accuracy of assessment results. Another study utilized the Apriori algorithm for correlation analysis of landslide evaluation factors and to screen the most relevant factors, employing the Random Forest model for susceptibility evaluation, resulting in assessments that closely matched actual landslide distributions. In a separate study, genetic algorithms and rough set analysis were used to screen landslide evaluation factors, while a BP neural network was employed for susceptibility evaluation, showcasing the enhanced accuracy of models with fewer evaluation factors. Models such as BP neural networks, support vector machines, Logistic regression, and Random forests have demonstrated their effectiveness in landslide susceptibility assessment. Moreover, comparisons between models like Random Forest, Logistic regression, Multilayer perceptron, and gradient enhancement tree have shown that the Random Forest model consistently yields the highest accuracy in landslide risk assessments. However, prior studies have sometimes overlooked the contribution of evaluation factors to landslide events during factor selection, potentially impacting the efficiency and accuracy of assessment results. Furthermore, data analysis methods often involve calculating the contribution rate of factors during screening, potentially leading to errors and outliers in the evaluation factor data derived from landslide events. Such errors may lead to the inadvertent elimination of significant factors during data analysis, affecting the accuracy of evaluation results.