Application of continuous restricted Boltzmann machine to detect multivariate anomalies from stream sediment geochemical data, Korit, East of Iran

Anomaly separation using stream sediment geochemical data has an essential role in regional exploration. Many different techniques have been proposed to distinguish anomalous from study area. In this research, a continuous restricted Boltzmann machine (CRBM), which is a generative stochastic artificial neural network, was used to recognize the mineral potential area in Korit 1:100000 sheet, located 15 km south of Tabas, South Khorasan Province (East of Iran). For this purpose, 470 geochemical stream sediment samples were collected from the study area and analyzed for 36 elements. In order to achieve the goal, in the first step, the robust factor analysis on compositional data was applied to reduce the data dimension and to limit the multivariate analysis by selecting the main components of mineralization. In this procedure, the third factor (out of 6) consisting of Cu, Pb, Zn, Sn, and Sb, related to the metallogenic properties, was considered as the input set in CRBM. In continuation, the CRBM structure with the best efficiency after trying different parameters was stabilized. High-identified error values or anomalies were exteracted using two different thresholds (ASC and ASE) after training with the whole data and reconstructing it by CRBM. The anomalies were then mapped. These indicated the promissing areas. The field studies and existing mining indices confirmly demonestrated the results obtained by CRBM.