Bagging Ensemble Based on Multi-Layer Perceptron Neural Network for Landslide Susceptibility Assessment

Landslide susceptibility assessment is the most appropriate method to mitigate the landslide threat for both populations and infrastructure. The development of accurate and reliable susceptibility models is therefore essential for sustainable development in landslide-prone regions. The present study proposes an implementation of the Bagging (BG) ensemble using the Multi-layer Perceptron Neural Network (MLP). The aim of this work is to reduce the variance of the MLP using the bootstrapping method of the BG ensemble. To estimate the potential of the ensemble, a fine-tuned individual MLP was implemented for benchmarking. A spatial database with 115 landslides and 12 causal factors was collected to implement the modelling. For separating training and test data, repeated K-fold cross-validation was used. The prediction capability of the models was estimated using the area under (AUC) the receiver operating characteristic (ROC) curve (AUC), the root mean square error (RMSE), the F1 and the accuracy scores. The results indicated that the BG ensemble outperformed the individual MLP in all the assessment measures. The BG-MLP model is more robust and generalizable, resulting in higher scores with a reduced standard deviation compared to the individual MLP. Therefore, the BG-MLP ensemble was successfully produced for the assessment of landslide susceptibility and demonstrated significantly improved results.