Erodibility of Nanocomposite-Improved Unsaturated Soil Using Genetic Programming, Artificial Neural Networks, and Evolutionary Polynomial Regression Techniques

Abstract

Genetic programming (GP) of four levels of complexity, including artificial neural networks of the hyper-tanh activation function (ANN-Hyper-Tanh), artificial neural networks of the sigmoid activation function (ANN-Sigmoid), evolutionary polynomial regression (optimized with genetic algorithm) (EPR), and intelligent techniques have been used to predict the erodibility of lateritic soil collected from an erosion site and treated with hybrid cement. Southeastern Nigeria and specifically Abia State is being destroyed by gully erosion, the solution of which demands continuous laboratory examinations to determine the parameters needed to design sustainable solutions. Furthermore, complicated equipment setups are required to achieve reliable results. To overcome constant laboratory works and equipment needs, intelligent prediction becomes necessary. This present research work adopted four different metaheuristic techniques to predict the erodibility of the soil; classified as A-7-6, weak, unsaturated, highly plastic, high swelling and high clay content treated with HC utilized in the proportions of 0.1–12% at the rate of 0.1%. The results of the geotechnics aspect of the work shows that the HC, which is a cementitious composite formulated from blending nanotextured quarry fines (NQF) and hydrated lime activated nanotextured rice husk ash (HANRHA), improves the erodibility of the treated soil substantially and consistently. The outcome of the prediction models shows that EPR with SSE of 1.6% and R2 of 0.996 outclassed the other techniques, though all four techniques showed their robustness and ability to predict the target (Er) with high performance accuracy.