Comprehensive analysis of soil electrical conductivity: an experimental and machine learning approach

Experimental test is conducted to find a relation between soil electrical conductivity with some soil properties. Five input parameters namely water content (w), dry density of soil (γd), degree of saturation (Sr), porosity (η) and voltage (V) are considered to compute electrical conductivity of soil (SEC). The effect of w, γd, Sr and η on SEC are analyzed. These datasets are used in soft computing technique to predict SEC. Five hybrid models namely ANN-GA, ANN-PSO, ANN-FA, ANN-GWO and ANN-MFO are used to predict SEC. To check the performance of these hybrid model, numerous statistical performance parameters (R2, a-10 index, VAF, LMI, RMSE, MAE, MAD and U95) are used. On the basis of these performance parameters, ANN-GA performs better (Due to higher value of R2, VAF, a-10 index and LMI and lower value of RMSE, MAE, MAD and U95) as compare to the other proposed model. The model's performance is also examined using rank analysis, regression curve, Williams plot, error matrix, objective function (OBJ) criterion, and Akaike information criterion (AIC). By applying all the above criteria, it has been observed that ANN-GA model outperforms than the other model to predict SEC. This was recognized to its maximum R2 = 0.9998 and the lowest RMSE = 0.0041 during the training phase, as well as R2 = 0.9975 and RMSE = 0.0112 during the testing phase. The model's reliability index (β) is calculated using the first-order second moment (FOSM) approach, and the result is compared to its actual condition. Additionally, a sensitivity analysis is carried out to ascertain the impact of each input parameter on the output (SEC).Based on the results, the hybrid model of ANN i.e., ANN-GA is potential to assist engineers to predict SEC in the design phase of high voltage buried power cables, ground modification techniques or in the field of agriculture.