Shallow landslide susceptibility assessment using a novel hybrid intelligence approach

We present a hybrid intelligent approach based on Naïve Bayes trees (NBT) and random subspace (RS) ensemble for landslide susceptibility mapping at the Bijar region, Kurdistan province (Iran). According to current literature, both NB and RS are machine learning techniques that have been rarely used for modeling of landslides. NBT is a relatively new decision trees-based algorithm in conjunction with Bayesian theories in building trees for classification, whereas RS is a relatively new ensemble framework with ability to improve performance of prediction models. In the hybrid approach, RS is used to generate subsets from the training data each subset is then used to construct a based classifier using NBT. For this purpose, a geospatial database for the study area was constructed that consisted of 111 landslide locations and 17 conditioning factors (slope degree, slope aspect, elevation above sea, curvature, profile curvature, plan curvature, stream power index, topographic wetness index, length-angle of slope, lithology, land use, distance to road, distance to fault, distance to stream, fault density, stream density, and rainfall). The database was used to construct and verify the proposed model. Performance of the model was evaluated using the receiver operating characteristics curve and area under the curve (AUC). The results showed that the proposed model performed well in this study (AUC = 0.886), and it improved significantly the performance of the NBT base classifier (AUC = 0.811). Overall, RS–NBT is promising which can be utilized for landslide susceptibility assessment in other landslide-prone areas.