<p>Different models and assessment factors, when applied to the same region, often yield divergent results. In order to study and propose a better model and its main controlling factors suitable for a specific region, reveal the geological law of hazards, and obtain better assessment results, we took the extreme rainfall-induced mass hazards event in the West Qinling Mountains in China as an example. According to the complexity of the model, four typical data-driven models, including the mathematical statistics and machine learning model were used to assessment the landslide susceptibility. Meanwhile, geomorphological process-oriented linear factors (e.g., hierarchically structured rivers and roads) and characteristic rainfall parameters considering hazard-causing mechanisms (i.e., effective rainfall during previous 3&#xa0;days and stimulated rainfall during previous 5&#xa0;h) were incorporated into the influence factor set. A variety of weight analysis methods were used to calculate the main controlling factors and analyse the key reasons affecting the accuracy of the model. The results show that ① the four data-driven models all had good prediction performances. The accuracies of these models were 0.767, 0.786, 0.852, and 0.874, respectively. Moreover, compared with the mathematical statistics, the prediction performance of the machine learning was significantly better. ② Based on our comprehensive calculations, the main controlling factors in the study area were the slope aspect, stimulated rainfall, etc. The main reason for the different accuracies of the prediction results of the different models is that the factors with low contributions and the linear factors are different. From the perspective of model evolution, this paper presents a targeted in-depth comparative study of different data-driven models. The results provide an effective theoretical reference for landslide susceptibility assessment in other regions.</p>