A comparative study of mutual information-based input variable selection strategies for the displacement prediction of seepage-driven landslides using optimized support vector regression

Artificial intelligence (AI) is becoming increasingly popular and useful for modeling landslide movement processes due to its advantages of providing excellent generalization ability and accurately describing complex and nonlinear behavior. However, the identification of key variables is a crucial step in ensuring robustness and accuracy in AI modeling, but thus far, little attention has been given to this topic. In the present study, mutual information (MI)-based measures are proposed for input variable selection (IVS) and incorporated into optimized support vector regression (SVR) for the displacement prediction of seepage-driven landslides. The performance of optimized SVR models with ten MI-based IVS strategies is compared. A typical seepage-driven landslide was chosen for comparison. The experimental results indicate that IVS-based optimized SVR can significantly improve predictions. When the variable-reduced inputs were input into the optimized artificial bee colony (ABC)-SVR model, the mean values of normalized root mean square error (NRMSE) and Kling-Gupta efficiency (KGE) decreased and increased by as much as 71.6 and 95.2%, respectively, relative to those for the base model with all candidates. Furthermore, the joint mutual information (JMI) and double input symmetrical relevance (DISR) criteria are recommended for IVS for seepage-driven landslides because they achieve the best tradeoff between accuracy and stability.

Automated summary

This study proposes an input variable selection method based on mutual information and incorporates it into an optimized support vector regression model for predicting the displacement of seepage-driven landslides. The experimental results show that the optimized model based on mutual information can significantly improve prediction accuracy and stability.