Studying the application of self organizing map (SOM) in stream sediment geochemical data clustering and comparing the results with compositional data dendrogram

Extensive development of the data mining methods via the artificial intelligence implementation and machine learning algorithms that the nature has inspired، have become an important challenge to the classical statistical analysis. Some constraints in the statistical assumptions are not made in these methods and just by defining the initial conditions and proper training، acceptable results can be achieved. Self organizing map (SOM) is a way that can unsupervisedly reduce the high dimensional complicated spaces to a 2 or 3D space and recognize the principal components without any difficult and almost impossible assumptions. Despite all the transformations applied to geochemical data، they intrinsically do not suit any statistical analysis and this is a serious factor for so many ifs and buts before analyzing the data. In this study، while introducing SOM as one of the most important approaches based on artificial intelligence، its usage in geochemistry has been demonstrated in a case study of stream sediment sampling carried out in Khusf geological 1:100000 sheet. Comparing the results of applying SOM on the compositionally scaled data and compositional univariate transformed data with exploratory compositional dendrogram of the data showed a favorable conformity of the dendrogram to the SOM clustering on the compositionally scaled data.