A hybrid approach of ANN and improved PSO for estimating soaked CBR of subgrade soils of heavy-haul railway corridor

The determination of subgrade/subsoil strength is one of the most important pavement design factors in transportation engineering, particularly for railways, roadways, and airport runways. The California bearing ratio (CBR) is often used to measure the strength and stiffness modulus of subgrade materials. This study presents a novel machine learning solution as an alternate approach for estimating soil CBR in soaked conditions. The present approach is an integration of an artificial neural network (ANN) and improved particle swarm optimisation (IPSO). According to experimental results during the testing phase, the proposed hybrid model, ANN-IPSO has achieved the highest predictive precision with root mean square error, RMSE = 0.0711 and mean absolute error, MAE = 0.0546. The findings of the proposed model are far superior to those of employed models including the conventional ANN, support vector machine, and group method of data handling. Six additional hybrid models of ANN and standard PSO (SPSO), PSO with time-varying accelerator coefficients, modified PSO, Harris hawks optimisation, slime mould algorithm, and colony predation algorithm were also constructed for a detailed comparison. Based on the outcomes, the newly created ANN-IPSO has the potential to be a new tool to estimate soaked CBR of fine-grained soils in civil engineering projects.