An approach based on socio-politically optimized neural computing network for predicting shallow landslide susceptibility at tropical areas

Abstract

A new hybrid model approach based on Imperialist Competitive Algorithm, a socio-politically optimization, and neural computing networks (ICA-NeuralNet) was developed and proposed in this study with the aim is to improve the quality of the shallow landslide susceptibility assessment at the Ha Long city area, Quang Ninh province. This area, which belongs to one of the three key economic regions of Vietnam, has a high urbanization speed during the last ten years. However, the landslide has been a significant environmental hazard problem during the last five years due to extreme torrential rainstorms. For this regard, a geographic information system (GIS) database was established, which contains 170 landslide polygons that occurred during the last five years and ten influencing factors. The database was used for training and validating the ICA-NeuralNet model. The results showed that the integrated model achieves high performance with classification accuracy rates of 82.4% on the training dataset and 78.2% on the testing dataset. Therefore, the ICA-NeuralNet is subsequently employed for generating a landslide susceptibility map of the study area, which greatly supports the land-use planning as well as hazard mitigation/prevention of local authority.