GIS-based Hybrid Machine Learning for Landslide Susceptibility Assessment in Thai Nguyen Province, Vietnam

Landslides are a natural hazard frequently occur- ring in mountainous areas and have been responsible for numer- ous fatalities and injuries. They were considered the primary cause of natural disasters leading to loss of life in Asia. In Vietnam, landslides have become increasingly common in recent years, particularly in mountainous regions, posing a direct and substantial hazard to life and property of human. To mitigate this risk, landslide susceptibility assessment is a valuable tool, but it requires continuous updating and further development with practical applications to address the complexity of landslide events. Given the importance of landslide susceptibility assessment, our work aimed to evaluate the susceptible to landslide utilizing hybrid machine learning (ML) techniques. Four ML models were employed for predicting landslide susceptibility, comprising SVM, RF, SVMMLP, and RFMLP. These models were constructed based on MLP (multilayer perceptron), RF (random forest), and support vector machine (SVM) algorithms. To develop the models, nine conditioning factors (slope, elevation, drainage density, relief degree of land surface, lithology, soil, land use land cover, rainfall, and NDVI) were collected relied on the correlation between historical landslide occurrences and specific geo-environmental conditions in the study area. The factors' quantitative importance has been determined by the difference in overall accuracy between the models with and without the conditioning factor. The models’ efficacy was evaluated through various statistical metrics such as sensitivity, accuracy, specificity, and the area under the receiver operating characteristic (ROC) curve (AUC). Elevated values of these parameters generally signify superior model performance. The findings indicated that the RFMLP model exhibited better performance, inslutrating by an accuracy of 85.33%, sensitivity of 85.71%, and specificity of 83.33%. Moreover, this model attained the highest AUC value at 86%, signifying its effectiveness in predicting landslide susceptibility. Consequently, the RFMLP model was employed to generate a landslide susceptibility map for CRWTN - our target site. This study has provided a practical approach for assessing landslide susceptibility utilizing hybrid machine learning models. Its significance lies in its practical applications for disaster risk reduction. The generated susceptibility map, particularly tailored to Thai Nguyen Province, will aid local authorities in effective land-use planning and disaster management. By utilizing advanced ML techniques and continuously updating these maps, authorities can proactively mitigate landslide risks, safeguarding lives and properties in regions prone to this natural hazard.