Bearing in mind that lack of data is a common problem in the study of porphyry copper mining exploration, our goal was set to identify the hidden patterns within the data and to extend the information to the data-less areas. To do this, the combination of pattern recognition techniques has been used. In this work, multi-layer neural network was used to estimate the concentration of geochemical elements. From 1755 surface and boreholes data available, analyzed by ICP, 70% was used for training, and the rest for testing. The average accuracy of estimators for 22 geochemical elements when using all data was equal to 75%. Based on validation, the optimal number of clusters for the total data was identified. The Gustafson-Kessel (GK) clustering was used to design the estimator for the geochemical element concentrations in different clusters, and the clusters were selected for estimation. The results obtained show that using GK, the estimator's average accuracy increase up to 84%. The accuracy of the elementsZn, As, Pb, Mo, and Mn with low accuracies of 0.51, 0.62, 0.64, 0.65, and 0.68 based on all data were developed to 0.76, 0.86, 0.76, 0.80, and 0.71 with the clustered data, respectively. The mean square error using all the data was 0.079, while in the case of hybrid developed method, it decreased to 0.048. There were error reductions in Al from 0.022 to 0.012, in As, from 0.105 to 0.025, and from 0.115 to 0.046 for S.