An Integrated Approach for Landslide Susceptibility Prediction using Decision Tree and Recurrent Neural Networks

Statistical and deep learning algorithms have been extensively employed in landslide susceptibility modelling due to their efficacy in mapping areas prone to landslides. This is especially applicable in extensive areas when geotechnical data is limited. Policymakers and land-use planners can significantly benefit from these data-driven methodologies for improved spatial planning to mitigate the risk of landslides. This research aims to assist deep learning models in addressing data preparation challenges to enhance their predictive accuracy about landslip probabilities. Data normalisation techniques and chi-square-based feature selection were employed to enhance efficiency. A novel hybrid model named BHyPreCis was developed to enhance the detection of intricate, non-linear patterns within the data. It is founded on the DBiLSRNGR architecture, which is integral to the BHyPreC framework. The testing results indicated that the model's predictive capability significantly enhanced following the integration of pertinent features. The proposed BHyPreCis model achieved an accuracy of 99.09% and an AUC of 99.43%, outperforming traditional statistical approaches. The results indicate that advanced deep learning algorithms can improve the prediction of landslip vulnerability. The study finds that hybrid modelling and meticulous data preparation are essential for generating reliable vulnerability maps, which subsequently enhance land-use planning and disaster risk mitigation strategies.