A novel soil liquefaction prediction model with intellectual feature extraction and classification

Abstract

This paper introduced a novel soil liquefaction prediction model with proposed features. Initially, pre-processing was carried out and then diverse improved features (chi-square features, relief features, technical indicators, and improved correlation-based features) were derived that were then predicted via SVM, LSTM, and DBN. The attained outputs from these classifiers were then predicted via optimized Bi-GRU that offered the final output. In particular, the Bi-GRU weights were tuned optimally via a novel AC-SSO model. Eventually, the primacy of the presented method was confirmed over existing schemes about varied measures. On analyzing the results, the presented scheme has achieved the slightest cost value (approximately 1.11), which was 1.35%, 0.36%, 0.63%, and 0.63% better than SSA, CSO, and GWO models. Thus, the excellence of the established approach was proven.