Detailed soil surveys involve costly and time-consuming work and require expert knowledge. Since soil surveys provide information to meet a wide range of needs, new methods are necessary to map soils quickly and accurately. In this study, multilayer perceptron artificial neural networks (ANNs) were developed to map soil units using digital elevation model (DEM) attributes. Several optimal ANNs were produced based on a number of input data and hidden units. The approach used test and validation areas to calculate the accuracy of interpolated and extrapolated data. The results showed that the system and level of soil classification employed had a direct effect on the accuracy of the results. At the lowest level, smaller errors were observed with the World Reference Base (WRB) classification criteria than the Soil Taxonomy (ST) system, but more soil classes could be predicted when using ST (7 soils in the case of ST vs. 5 with WRB). Training errors were below 11% for all the ANN models applied, while the test error (interpolation error) and validation error (extrapolation error) were as high as 50% and 70%, respectively. As expected, soil prediction using a higher level of classification presented a better overall level of accuracy. To obtain better predictions, in addition to DEM attributes, data related to landforms and/or lithology as soil-forming factors, should be used as ANN input data

Most common machine learning (ML) algorithms usually work well on balanced training sets, that is, datasets in which all classes are approximately represented equally. Otherwise, the accuracy estimates may be unreliable and classes with only a few values are often misclassified or neglected. This is known as a class imbalance problem in machine learning and datasets that do not meet this criterion are referred to as imbalanced data. Most datasets of soil classes are, therefore, imbalanced data. One of our main objectives is to compare eight resampling strategies that have been developed to counteract the imbalanced data problem. We compared the performance of five of the most common ML algorithms with the resampling approaches. The highest increase in prediction accuracy was achieved with SMOTE (the synthetic minority oversampling technique). In comparison to the baseline prediction on the original dataset, we achieved an increase of about 10, 20 and 10% in the overall accuracy, kappa index and F‐score, respectively. Regarding the ML approaches, random forest (RF) showed the best performance with an overall accuracy, kappa index and F‐score of 66, 60 and 57%, respectively. Moreover, the combination of RF and SMOTE improved the accuracy of the individual soil classes, compared to RF trained on the original dataset and allowed better prediction of soil classes with a low number of samples in the corresponding soil profile database, in our case for Chernozems. Our results show that balancing existing soil legacy data using synthetic sampling strategies can significantly improve the prediction accuracy in digital soil mapping (DSM). Highlights: Spatial distribution of soil classes in Iran can be predicted using machine learning (ML) algorithms. The synthetic minority oversampling technique overcomes the drawback of imbalanced and highly biased soil legacy data. When combining a random forest model with synthetic sampling strategies the prediction accuracy of the soil model improves significantly. The resulting new soil map of Iran has a much higher spatial resolution compared to existing maps and displays new soil classes that have not yet been mapped in Iran.