Enhancing the Accuracy of the REPTree by Integrating the Hybrid Ensemble Meta-Classifiers for Modelling the Landslide Susceptibility of Idukki District, South-western India

Idukki district, situated in the Western Ghats of Peninsular India, is one of the high landslide susceptible zones with frequent landslide occurrences during monsoon. Though plentiful studies have been carried out, this study aims to utilize a model based on a decision tree, called the REPTree algorithm, and its ensemble models such as Random Subspace-REPTree, AdaBoost-REPTree, MultiBoost-REPTree, and Bagging-REPTree to identify the landslide susceptibility zones. For the susceptibility modeling, twelve landslide conditioning factors were used. The receiver operating characteristic (ROC) curve, precision, mean absolute error (MAE), and root-mean-square error (RMSE) were used to validate the results. The results of the validation revealed that all models are effective (precision: > 0.840, AUC: > 81.00%, RMSE: < 0.285, and MAE: < 0.205) in identifying the landslide susceptibility of the Idukki district. The RSS-REPTree model achieved the highest accuracy (precision: 0.879, AUC: 88.82%, RMSE: 0.153, and MAE: 0.133) among the five models. Thus, the identified landslide susceptible zones will help the land-use planners and officials of the emergency management department in implementing mitigation measures that are effective for this terrain and other terrain having similar geological, geomorphological, and climatic conditions.

Automated summary

This study utilizes various models and their ensemble models to identify landslide susceptibility zones in Idukki district. The results show that all models have good performance, and the identified zones will help in implementing effective mitigation measures.