Landslide susceptibility assessment using novel hybridized methods based on the support vector regression

Landslide susceptibility assessment is a complex task due to the multitude of causative factors, spatial variability, data availability, modeling uncertainty, and validation issues. This study addresses these challenges by proposing two predictive models that hybridize support vector regression (SVR) with two evolutionary algorithms: grey wolf optimizer (GWO) and cuckoo search algorithm (CSA). These models were developed using an extensive geospatial database from northern Iran. Over the training phase, the basic predictive model, developed using SVR, was enhanced by incorporating the GWO and CSA algorithms, resulting in the development of two hybridized models: SVR-GWO and SVR-CSA. Over the validation phase, the performance and effectiveness of each hybridized model were compared to the standalone SVR using several metrics. Compared to the standalone SVR model, the hybridized models demonstrated significant improvement in generalization and predictive abilities. Despite non-significant difference between the performances of the SVR-GWO and SVR-CSA models, the SVR-GWO model demonstrated superior performance. This could be attributed to the GWO's capabilities, which included generating a variety of solutions, demonstrating robustness against noise and outliers, achieving faster convergence speed, and effectively interacting with SVR. This study highlighted that utilizing intelligence hybridized models can significantly enhance the balance between accuracy, robustness, and objectives compared to single models. This finding holds significant implications for ecological engineers tasked with designing and implementing solutions to mitigate the impact of shallow landslides on the environment and human communities. The predictive models developed in this study serve as valuable tools for these engineers, enabling them to identify high-risk areas and implement preventative measures. This interdisciplinary approach, which combines machine learning, optimization algorithms, and ecological engineering, highlights the potential for pioneering solutions in tackling complex environmental challenges, thereby standing as a testament to the power of innovation in driving progress in landslide susceptibility assessment.