Evaluating the factors affecting landslides using machine learning algorithms (case study: the catchment area of Karun-3 Dam, Iran)

Landslides are among the phenomena associated with environmental impacts and human and financial losses worldwide. Investigating environmental issues such as landslides and preparing hazard maps are essential for managers and planners. This study examines and models landslides in the catchment area of Karun-3 Dam located in Khuzestan province, Iran, using six machine learning algorithms, including Random Forest (RF), Boosted Regression Tree (BRT), Generalized Aggregate Model (GAM), Support Vector Model (SVM), Classification and Regression Tree (CART), and Generalized Linear Model (GLM). Thirteen independent parameters were identified as the main parameters. Then, their correlation and effects were examined using 284 old landslides, and machine learning models were validated using efficiency, sensitivity, and accuracy indicators. The validation results showed that although all the models used have sufficient accuracy, the RF model (AUC = 0.982, Efficiency = 0.943) has more accuracy than the other five models. Also, the impact of different factors on landslide generation in various models is not the same. In general, the significance of the mentioned parameters is in the range of 0.043 and 0.160. Comparing the results of different models using a non-parametric test shows more similarities between the models used. In general, the results of various models show that the risk of landslides is generally higher on the steep banks of rivers, in the vicinity of lakes, dams, and roads, and especially in lands with soft lithology such as marl. This fact shows us the influence of anthropogenic factors and natural factors simultaneously.