Machine learning based soil erosion susceptibility prediction using social spider algorithm optimized multivariate adaptive regression spline. (November 2020)

• Record Type: Journal Article
• Authors: Vu, Dinh Tuan; Tran, Xuan-Linh; Cao, Minh-Tu; Tran, Thien Cuong; Hoang, Nhat-Duc
• Is Part Of: Measurement. Volume 164(2020)
• Publication Date: 2020-11
• Subjects: Soil erosion -- Multivariate adaptive regression splines -- Social spider algorithm -- Hybrid machine learning; Weights and measures -- Periodicals; Measurement -- Periodicals; Measurement; Weights and measures; Periodicals; 530.8
• DOI: 10.1016/j.measurement.2020.108066
• Languages: English
• ISSNs: 0263-2241

Abstract

Highlights: Propose a machine learning based soil erosion susceptibility prediction. Multivariate Adaptive Regression Splines (MARS) is used for data analysis. Social Spider Algorithm (SSA) is employed for model optimization. An integration of SSA and MARS is proposed. The proposed approach achieves a high predictive accuracy (96%). Abstract: This study proposes an advanced data-driven method which relies on the Multivariate Adaptive Regression Splines (MARS) machine learning and Social Spider Algorithm (SSA) metaheuristic for predicting soil erosion susceptibility. The MARS is employed to infer a decision boundary that separates the input data space into two distinctive regions of 'erosion' and 'non-erosion'. Meanwhile, the SSA metaheuristic is aimed at optimizing the MARS performance by automatically fine-tuning its hyper-parameters. The proposed SSA optimized MARS method, named as SSAO-MARS, is trained and validated by a set of 236 samples of soil plot conditions associated with their corresponding erosion status. The research finding shows that the newly developed SSAO-MARS can attain good predictive outcomes with classification accuracy rate of roughly 96%. Therefore, the newly developed model can be a useful tool to assist land management agencies.