Development of integrated deep learning and machine learning algorithm for the assessment of landslide hazard potential

In mountainous regions subjected to landslides, susceptibility mapping of these geohazards is necessary for averting and alleviating perilous dangers. The present study applies an integrated methodology for assessing landslide susceptibility of northern Pakistan (Mansehra and Muzaffarabad districts). Three orthodox machine learning (ML) classification techniques, including support vector machine (SVM), logistic regression (LR), and random forest (RF), are integrated with convolutional neural network (CNN) used. For training and testing of the models, spatial datasets consisting of 3251 sites of historical slopes are used in a ratio of 70:30. Initially, a total of 16 influencing factors for landslide modelling were established. The training dataset specifically constructs three hybrid models CNN-SVM, CNN-LR, and CNN-RF. Then, final susceptibility maps (LSMs) will be built using these trained models. These models will be implemented. For having a comparison, the LSMs are also prepared using the considered ML models individually. In the end, multiple statistical methods are used to validate and compare the performance of these models. The results of the analysis have revealed the efficiency of applying the projected ML models by combining them with the CNN technique. Therefore, in other sensitive regions with comparable geo-environmental conditions, the future hybrid designs can be used effectively for landslide susceptibility studies.

Automated summary

This study assesses landslide susceptibility in northern Pakistan using a combination of methods including machine learning techniques such as support vector machine, logistic regression, random forest, and convolutional neural network. By building hybrid models and susceptibility maps, the efficiency of combining ML models with CNN technique is demonstrated and can be effectively applied in other sensitive regions with similar geological conditions.