The diagnosis of saline soils requires the analysis of electrical conductivity in saturated soil paste extract. Its analysis is expensive, tedious, and highly time-consuming, therefore, commercial laboratories analyze the aqueous extract in a 1:1 ratio and then transform the value into saturation extract using equations. The research aimed to calibrate a statistical learning method to predict the electrical conductivity adapted to Peruvian conditions. For this, we apply different models from highly interpretable to black-box, such as multiple linear model, generalized additive models, Bayesian additive regression tree, extreme gradient boosting trees, and neural networks. In general, the models with beast predictive power were neural network and extreme gradient boosting trees, and the beast interpretable was Bayesian additive regression trees. The generalized additive models present the best balance between prediction power and interpretability with low application on extremely salty soils.