Deep learning algorithms based landslide vulnerability modeling in highly landslide prone areas of Tamil Nadu, India

Landslide is a common hazard in Tamil Nadu’s Nilgiri district of. While much work on landslide susceptibility have been done worldwide, understanding society’s vulnerability to landslides, considering house structure and socio-economic conditions, remains lacking. This research presents landslide vulnerability mapping using advanced computing deep learning models in the Nilgiri district, India. Compared to traditional ML techniques, the deep learning neural network (DLNN) architecture demonstrates greater accuracy, particularly when dealing with more samples or significant amounts of big data. Although the standardized characteristics of multi-layer NNs are widely known, the key benefit of DL is its organized method for training DLNN-layer organizations how to govern themselves. Therefore, one deep learning neural network and three conventional machine learning models i.e., MLP classifier and RBF neural network were opted. A total of twenty-eight physical, climatological, hydrological and socio-economic factors were considered to produce socio-economic and relative landslide vulnerability maps. Multi-collinearity diagnosis was performed to select the appropriate factors. Several physical as well as human related factors are highly important for making the area vulnerable to land-slide. To justify the vulnerability maps, several statistical methods were applied. The best model DLNN, with an area under the curve of 89.07%, shows that 43.31%, and 37.72% of areas are highly to very-highly vulnerable to landslides. The framework presented in this work establishes an ideal link between human activities and landslide vulnerability, aiding planners in making informed decisions for landslide management.