Multivariate Statistics and Machine Learning Techniques as Tools for Exploration Targeting in Parts of the Tanzania Craton: Insights from Stream Sediments Geochemistry

Geochemical exploration has been an indispensable approach to mineral exploration and has been responsible for significant mineralization discoveries across the globe. With the increasingly difficult exploration targeting, there is a need for an add-on in the geochemical data processing techniques that has the qualities of enhancing exploration targeting even with the slightest of geochemical signatures. In recent years, machine learning methods coupled with multivariate statistical methods have gradually proven to be a good add-on to the geochemical data processing techniques, which are yielding positive exploration targeting results. Although similar studies aimed at enhancing exploration targeting have been conducted in some parts of the Tanzania Craton (TC) using the same approach, these earlier works did not include the sensitivity analysis of the mineralization characteristics within the TC. Hence, this study seeks to trace the potential source(s) of Au-Pt-Pd in stream sediments, their pathfinder elements, and the most sensitive elements to their occurrence in the study area. The study adopted multivariate statistics together with regression modeling, artificial neural network (ANN), and sensitivity analysis methods. The elemental association of Sr, Ba, Nb, and Rb suggests felsic igneous rocks, and Co, Ni, V, and Cr indicate mafic rock sources of the Au-Pd-Pt. The most sensitive elements to the Au occurrence in the area are Zn, Bi, Sr, Ca, K, Sn, Y, and Nb, while those for Pd are Au and Pt. The prediction model for Au performed better with RE values of 0.787 and 0.576 in training and testing stages, respectively. The geological complexity of the TC might have accounted for the variable pathfinder elements of Au within the different Au-bearing catchments.