Implementing an explored advanced and integrated deep random forest learning-based model to monitor the enhanced landslide susceptibility mapping

The landslides susceptibility analysis is used to identify the susceptible area for landslides occurrence which utilized in geo-hazard analysis and managements. AI techniques are widely used by the practicing engineer to solve a whole range of hitherto intractable problems, so, providing accurate susceptibility mapping with advance AI methods and machine learning helps to reduce the land-sliding consequences which are one of the most important steps in urban and hazard managements in real world engineering application. The presented study used advanced random forests algorithm to provide the susceptibility maps of landslide hazard in Fars province in Iran. In this regard, a landslide inventory database is prepared based on totally 352 historical landslides and 5 main triggering factors (e.g., morphologic, geologic, climatologic, seismicity, and human-activity parameters). This database was used to prediction process by machine learning method that trained by 70% and validated by 30% of the main database. The model was analysed based on confusion matrix, loss function and validated via overall accuracy with receiver operating characteristics (ROC) curve. The result shows that random forests algorithm reached the considerable overall accuracy (AUC = 0.944). Additional, by estimating the error rates concluded mean absolute error (MAE), mean squared error (MSE), and root-mean-square error (RMSE) on both testing and training sets; it’s appeared that the model was operated with remarkable accuracy (AUC > 0.9) that indicate the high- capability of the predictive model in landslides susceptibility assessment.