A geochemical modeling to predict the different concentrations of REE and their hidden patterns using several supervised learning methods: Choghart iron deposit, bafq, Iran

• We perform extensive experiments to examine the influence of various rocks' compositions on distribution patterns of REE.
• Decision tree has been selected to achieve a more clear comprehension of the REE’ behavior.
• Some REE distributions are reasonably predictable, and others are too irregular to be modeled.
• It is worth noting that classification rules can be additional factors illustrating the REE behavior.
• The applied technique can be extended in similar projects with a considerable saving in time and cost.

The Bafq mining district hosts some Kiruna-type iron oxide-apatite (IOA) deposits which are commonly formed as a result of the multistage interaction of hydrothermal-magmatic processes within the Early Cambrian volcano-sedimentary sequence. Rare earth elements (REEs) are potentially concentrated under different physicochemical conditions in IOA deposits. Choghart orebody is one of the main magnetite-apatite deposits in the region. A wide range of primary and secondary geological events have affected the Choghart deposit, causing the behavior of REEs to vary in different zones. This study proposes a suitable exploration technique using various classification methods to identify the different concentrations of REE and their hidden patterns. To provide the required data, a systematic lithogeochemical sampling was performed in the north and NE of the Choghart orebody. The REE contents of samples were transformed into discrete values as distinct classes based on the results of clustering analysis. All possible combinations of features, being the geographical location and the major oxide composition of samples, were selected as subsets of predictors in every classification method. For each REE, 455 prediction models were constructed using these predictors. The performances of the classification methods were evaluated by error criteria with regard to all cases. Having the least amount of errors, the decision tree method was selected as the most suitable method. Based on decision tree results, the best subsets of predictors were chosen for each element. The existence of a significant relationship between the distribution patterns of each REE and the related predictors was assessed by its prediction errors. The assessment illustrated that some REE are reasonably predictable, and others are too irregular to be modeled. The extracted classification rules describe the geochemical relationships among the most important factors influencing the different concentrations of REE in the Choghart orebody. These results can be extended to other similar deposits to predetermine some REE-enriched zones based on major elements analysis. Merely by employing such techniques in REE exploration projects, a great savings in time and cost will be affected.