Objectives The influence of environmental factor classification interval and machine learning model on modeling results in landslide susceptibility evaluation modeling cannot be ignored. In order to explore the influence of these two factors on the evaluation results of landslide susceptibility.

Methods The evaluation index system is constructed by weighting the environmental factors based on the subjective and objective weighting method, and then the influence of different continuous variable factor classifications on the accuracy of landslide susceptibility evaluation results is explored by using the GeoDetector. Then, the random forest model, the gradient boosting model and the neural network model optimized by genetic algorithm are used to study the landslide susceptibility.

Results The results show that: the maximum area under curve value calculated by the partition combination with the highest correlation degree with the disaster obtained by the GeoDetector is 0.886.This finding indicates that the method can obtain the optimal classification interval and can effectively improve the accuracy of the susceptibility evaluation results. In the evaluation of susceptibility, the random forest model has been demonstrated to be the most effective, with a 9.7% and 9.6% increase over the gradient limit lifting model and the neural network model optimised by genetic algorithm, respectively.

Conclusions The optimal classification interval of environmental factors, as determined by GeoDetector, is reasonable, and the random forest model is both efficient and accurate in its evaluation of landslide susceptibility.