Advanced predictive modelling of electrical resistivity for geotechnical and geo-environmental applications using machine learning techniques

• Published in: Journal of applied geophysics Vol. 231; p. 105557
• Main Authors: Kundu, Soumitra Kumar, Dey, Ashim Kanti, Sapkota, Sanjog Chhetri, Debnath, Prasenjit, Saha, Prasenjit, Ray, Arunava, Khandelwal, Manoj
• Format: Journal Article
• Language: English
• Published: Elsevier B.V 01.12.2024
• Subjects: Artificial neural networks; Electrical resistivity; Extreme Gradient Boosting; Support vector regression; Support vector regression; Extreme Gradient Boosting; Electrical resistivity; Artificial neural networks
• ISSN: 0926-9851
• DOI: 10.1016/j.jappgeo.2024.105557

Electrical Resistivity (ER) is one of the best geophysical methods for subsurface investigation, especially for geotechnical and geo-environmental studies. Being non-invasive, economical and rapid, this method is highly preferable to geotechnical engineers for continuous evaluation of soil properties along the resistivity profile. Numerous studies have been conducted to correlate the subsurface properties with the ER. However, most of the studies consider a single input variable, which is correlated with the resistivity values using some conventional regression analyses. Very few studies have been conducted to obtain the resistivity value with multiple input parameters, like unit weight, temperature, porosity, moisture content, etc. Since, the soil parameters have a combined effect on resistivity, hence, correlations between the resistivity and the multiple input parameters are urgently required for a better and more reliable result. Moreover, the non-linear properties of soil make the task more complicated. To fill up this research gap, in the present study, 2772 ER tests were conducted using seven different types of soil with different combinations of temperature, density, and water content. Using this database, a Support Vector Regression (SVR), Artificial Neural Network (ANN) model and Extreme Gradient Boosting (XGB) were developed for prediction of ER. It has been understood that all the models are acknowledged as trustworthy data modelling tools. However, the XGB model performs better with an R2 of 0.99 during the training and testing phase. Further, a parametric study was also done to determine, how each input parameter affects the ER. An error analysis was also performed to see the consistent discrepancy between the experimental and projected values of ER. The outcomes validate the robustness of the XGB model, indicating that it can serve as a substitute method for ER prediction. •2772 Electrical Resistivity (ER) tests on 7 soil types using temperature, density, and water content data.•Explored multiple input parameters to develop different soft computing predictive models.•Extreme Gradient Boosting model excels with R2 of 0.99 in ER prediction.•Parametric study shows individual impacts of temperature, density, and moisture on ER.•Error analysis confirms alignment between experimental and predicted ER values.