Prediction of landslides by machine learning algorithms and statistical methods in Iran

English. peer reviewed

English. The present study aims to compare the performance of artificial neural networks (ANN), statistical methods, and machine learning in predicting the location of landslides in Marivan and Sarvabad cities in Kurdistan province of Iran. 16 factors influencing landslide occurrence are first chosen, such as aspect and elevation, plan curvature and profile curvature, slope degree, land use and geology, distance from faults, rivers and roads, rainfall, NDVI, SPI, STI, TRI and TWI. Then, the correlation between the factors affecting landslides and the ANN model was estimated through the FR method with 0.83% accuracy of the variable significance in the modeling, and the landslide sensitive zones were mapped in five classes (very high, high, medium, low, and very low). The results of the modeling analysis indicated that more than 70% of the study areas have high and very high sensitivity to the occurrence of amplitude movements. The accuracy of the prediction performance of the models used on the data (test-data and train-data) has been obtained 0.82, 0.83 and 1 with ANN, KNN and ADTree, respectively. Then, the efficiency of the models for prediction operations were evaluated with the classification matrix algorithm and the variables’ accuracy, sensitivity, and specificity. The findings of this stage showed that efficiency of the KNN was %74 and in ADTree algorithm was equal to 1. Therefore, results showed that in the comparison between the models used the ADTree algorithm has high accuracy and capacity to classify and predict the areas sensitive to landslides. Therefore, results showed that in the comparison between the models used the ADTree algorithm has high accuracy and capacity to classify and predict the areas sensitive to landslides. Besides, it can be utilized as the most efficient model in landslide disaster management.