Comparing compositional multivariate outliers with autoencoder networks in anomaly detection at Hamich exploration area, east of Iran

A comparison between both methods has been made applying them on lithogeochemical samples of Hamich area in Southern Khorasan, East of Iran. The area was explored in details some years ago and veinlets of galena-sphalerite-pyrite at depth, based on the outcrops of Cu-Pb, were verified by additional core drillings. We used its final report to validate the results of both methods. They showed that the two completely different methods could get the same acceptable targets. However the CoDa approach needs less parameters and shows which elements are responsible for the anomalies.