Predicting slope failure with intelligent hybrid modeling of ANFIS with GA and PSO

In geotechnical engineering, soil slopes are crucial in various civil engineering projects, including highways, embankments, dams, and excavations. Understanding the behavior of soil slopes is essential for designing stable and safe structures. Combining different soft computing (SC) models can provide more robust slope stability predictions. This paper employs two hybrid computational algorithms to make accurate slope stability predictions. In this research project, the adaptive neuro fuzzy inference system (ANFIS) model is optimized by two novel meta-heuristic optimization algorithms (MOAs): genetic algorithm (GA) and particle swarm optimization (PSO). To this end, slope inputs are taken from a literature survey consisting of 206 input datasets for the training and testing of models. Eleven statistical indices have been evaluated for assessing the performance of proposed hybrid models, along with evaluating rank analysis. ANFIS, ANFIS-GA, and ANFIS-PSO outcomes from the suggested models have R2 values of 0.6783, 0.7624, 0.7378 during training, 0.6684, 0.8143, and 0.7013 during testing. Also, the ANFIS-GA hybrid model yielded error matrices such as RMSE, MAE, and MSE with values of 0.1217, 0.0912, and 0.0148 in training and 0.12570, 0.0968, and 0.1391 in testing; in contrast, the ANFIS PSO model yielded values of 0.1264, 0.0902, 0.016 in training, and 0.1591, 0.1170, 0.1290 in testing; the ANFIS model yielded values of 0.1345, 0.1127, 0.0172 in training, and 0.1642, 0.1267, 0.1391 in testing. The regression plot was analyzed to compare the predicted value with the actual one. In the present paper, the Metropolis Hastings MCMC sampling method has been introduced to establish the relationship between the inputs, which is slope height (H), slope angle (alpha), cohesion (c), pore water pressure ratio (Ru), unit weight (Upsilon), angle of internal friction (phi), and output reliability of slopes. A sensitivity analysis was also performed to determine which variable affects the reliability of soil slope more. After that, comparing hybrid models with the ANFIS model notified the engineers and researchers that the model best predicts slope failure for extensive observations.