Abstract
Recently, machine learning models have acted as effective tools for slope stability analysis. But due to the crucial significance of this issue, reaching a reliable accuracy is necessary. This study investigates the efficacy of teaching–learning-based optimization (TLBO) for tuning two well-known predictive models, namely artificial neural network (ANN) and adaptive neuro-fuzzy inference system (ANFIS), applied to the prediction of the factor of safety for a cohesive soil slope-footing system. The conventional versions are also used as benchmarks. A finite element analysis provided the required dataset. During the implementation of the models, it was shown that both hybrid models show a high sensitivity to the population size of the TLBO. A qualitative assessment of the results showed that all four models yield a valid prediction (testing Pearson correlation indices are 0.9966, 0.9953, 0.9972, and 0.9970). A comparison between the prediction results revealed notable improvements for both models. In this relation, the root-mean-square error of the ANN and ANFIS were 0.4958 and 0.5867, which after applying the TLBO were reduced to 0.4511 and 0.4629, respectively. As for the mean absolute error, the values decreased from 0.3848 and 0.4160 to 0.3291 and 0.3640. Accordingly, ANN-TLBO was the strongest model, followed by ANFIS-TLBO, ANN, and ANFIS. The use of ANN-TLBO can thus be recommended for dependable design of soil-footing systems in areas of critical slopes. The formulation of this model is presented herein for its convenient use in practical projects.