Two novel neural-evolutionary predictive techniques of dragonfly algorithm (DA) and biogeography-based optimization (BBO) for landslide susceptibility analysis

Due to the wide application of evolutionary science in different engineering problems, the main aim of this paper is to present two novel optimizations of multi-layer perceptron (MLP) neural network, namely dragonfly algorithm (DA) and biogeography-based optimization (BBO) for landslide susceptibility assessment at a study area, West of Iran. Utilizing 14 landslide conditioning factors, namely elevation, slope aspect, plan curvature, profile curvature, soil type, lithology, distance to the river, distance to the road, distance to the fault, land cover, slope degree, stream power index (SPI), and topographic wetness index (TWI) and rainfall as the input variables, and 208 historical landslides as target variable, the required spatial database is created. Then, the MLP is synthesized with the mentioned algorithms to develop the proposed DA-MLP and BBO-MLP ensembles. Three accuracy criteria of mean square error, mean absolute error, and area under the receiving operating characteristic curve are used to evaluate the performance of the models and also to develop a score-based ranking system. As the first outcome, the application of the DA and BBO metaheuristic algorithms enhances the accuracy of the MLP. Moreover, referring to the calculated total ranking scores of 6, 14, and 16, it was revealed that the BBO performs more efficiently than DA in optimizing the MLP.