The goal of this study is the preliminary mapping of P-Li-Be contents related to pegmatites fields from the Eastern Brazilian Pegmatite Province (EBPP) at the Conselheiro Pena pegmatite district of Minas Gerais State. The study was based on machine-learning tools, gathering air-borne gamma-ray spectrometric and magnetometric data, Sentinel-2 MSI data, numerical terrain models, and field data (47 samples). Based on the relation between the element’s contents and covariates (remote sensing and topographic data) statistical models were built to represent the spatial variability of Li, Be and P₂O₅contents. Posteriorly, the resulting maps were combined through cluster analysis (k-means). The model’s validation was based on 63 control samples from the Geological Survey of Brazil (CPRM) not used at the model’s training procedures. In general, the Sentinel, radiometric and topographic data were related to all elements, however magnetometric data showed low relevance. Better performance to predict all three elements were obtained through tree-based models, in which Random Forest models showed better performance based on statistical accuracy indexes and generalization for the entire area. The cluster analysis identified five classes based on a combination of Li, Be, P₂O₅ contents, and based on this approach was possible to identify areas with the possibility to prospect LCT-Type pegmatites bodies. Zone 3 showed better concordance with control samples (43%) and the geological map where embedding rocks correspond to schist, gneisses, and granites with mineralized pegmatite bodies. Based on preliminary analysis presented in this paper the approach showed to be useful to reduce the initial area to less than 30% that are prospectable to pegmatite fields. Detailed studies should be carried out to understand better the behavior of variability in zones 2, 3, and 4, where special attention should be paid to the structures that condition the pegmatite’s emplacement.