Classifying Soil Sulfate Concentration Using Electrical Resistivity Imaging and Random Forest Algorithm

Assessing the levels of sulfate concentration and the extent of zones prone to sulfate-induced heaving is essential in designing appropriate stabilization and modification methods. However, due to spatial and temporal variations of sulfate minerals, conventional methods of measuring sulfate that provides point-specific data lead to misrepresentation of sulfate concentration. This study proposes an approach for identifying sulfate concentration levels and distributions based on electrical resistivity imaging (ERI) and random forest (RF) algorithm. Laboratory experiments were carried out on several soil samples based on a factorial design to assess the effects of sulfate concentration and water content on the electrical resistivity of soils. The sulfate concentrations of soil samples were modified by adding calcium sulfate in 1,000 ppm increments to represent sulfate concentrations ranging from 0 to 12,000 ppm. A random forest model was then trained using collected data from the laboratory (382 observations) to classify the sulfate concentrations into three levels: low (below 3,000 ppm), medium (between 3,000 ppm and 8,000 ppm), and high (above 8,000 ppm). Facing an imbalanced classification problem (i.e., one class has more observations than the others), a synthetic minority over-sampling technique (SMOTE) was used to remedy the imbalanced class distribution and increase the predictive performance of the trained model. This study showed that the random forest classifier could identify the levels of sulfate concentration using electrical resistivity and moisture content with an accuracy of 68.8%. The results showed that the random forest classifier performs better in determining the true observations from all observations in moderate sulfate concentration levels with a recall of 79% than low and high sulfate concentration levels. The proposed approach could help pavement engineers in construction and treatment decisions by determining the spatial variability of sulfate concentration levels and identifying areas with a potentially high risk of sulfate-induced heaving. Making well-informed decisions can prolong pavement service life and lower maintenance and rehabilitation costs.