Bio-inspired Predictive Models Development for Strength Characterization of Cement Deep-Mixed Plastic Soils

This paper utilizes various artificial intelligence models to predict the experimental results of the deep-mixing technology for ground improvement and stabilization. A total of 192 unconfined compression strength laboratory experiments were conducted on specimens taken from Khuzestan, Iran, to compare the strength of the soil before and after the treatment with cement. In this research, 144 sets of experimental data, constituting 75% of the total, were used for training, while 48 sets, equivalent to 25% of the experimental data, were utilized for both testing. Different artificial intelligence methods including artificial neural networks, hybrid artificial bee colony-artificial neural networks, combinational group modeling of data handling, and gene expression programming were used. To evaluate the performance of each method, mean squared error, root mean squared error, mean absolute percentage error, mean absolute error, linear correlation coefficient, and coefficient of determination was calculated for each method. Based on the performance analysis, the hybrid artificial bee colony-artificial neural network algorithm outperformed other methods with an R² calculated as 0.9969 and 0.9952, respectively in training and testing. The R² values during training for ANN, GMDH, and GEP are 0.993, 0.983, and 0.96, respectively. Likewise, during testing, the R² values for ANN, GMDH, and GEP are 0.992, 0.978, and 0.953, respectively. The results demonstrated significant agreement between artificial intelligence predictive models and experimental data. Furthermore, this paper provides robust and cost-effective models with a “closed-form solution” for predicting the strength of stabilized soils by the deep mixing technique. The closed-form equation presented in this study, which is derived from the group method of data handling combinatorial algorithm and gene expression programming models, is more intuitive for engineers to apply.