Bio-inspired optimization of weighted-feature machine learning for strength property prediction of fiber-reinforced soil

The fiber-reinforcement of soil is an effective and reliable ground improvement technique for increasing the strength and stability of soil for various purposes (including retaining structures, embankments, foundations, slopes and pavements). Numerous scholars have developed methods to identify factors that influence the shear strength and to predict the peak friction angle of fiber-reinforced soil (FRS). The accuracy of theoretical and empirical models for predicting the shear strength (peak friction angle) of FRS is questionable because of the difficulty of using these simplified models to describe the complex mechanism of soil-fiber interaction. Solutions to this problem require ever-increasing predictive accuracy, and ML-based methods have been confirmed to provide potential solutions to real-world engineering problems. Therefore, this study develops weighted-feature least squares support vector regression (WFLSSVR) that is optimized by a novel metaheuristic algorithm, jellyfish search (JS) algorithm, to predict the peak friction angle of FRS. Analytical results demonstrate that JS-WFLSSVR outperforms baseline, ensemble, and hybrid machine learning models as well as empirical methods in literature. Notably, analysis of the weight values that were obtained by JS-WFLSSVR enables the identification of new feature combinations that provide much higher accuracy than current models. Therefore, the JS-WFLSSVR model not only significantly provides better predictive accuracy than methods in the literature; it is also a good feature selection method, and can help geotechnical engineers in estimating the shear strength of FRS. Geotechnical engineers can use the proposed model to predict the shear strength and control the quality of FRS structures.