Provenance studies of Au-bearing stream sediments and performance assessment of machine learning-based models: insight from whole-rock geochemistry central Tanzania, East Africa

The source of clastic sediments generally, can be traced to their source through provenance studies using the whole rock geochemistry of clastic sediments. However, the provenance of the Au-bearing stream sediments within the central parts of Tanzania is yet to be deciphered. Hence, in this study, to enhance exploration targeting, the source of the Au-bearing stream sediments was characterized using whole-rock geochemistry. The performance of linear regression (LR), decision tree (DT), and polynomial regression (PR) models as prediction models for the Au mineralization in the area, were also compared as additional Au exploration techniques worth exploring in the area. The weathering condition proxies, CIA, ICV, CIW, and PIA as well as discriminant diagrams suggest weakly to intensely weathered sediments. The values of SiO₂/Al₂O₃ and K₂O/Al₂O₃ are indicative of felsic source rocks rather than compositional maturity due to sediments reworking. From Th/Cr, Cr/Th, Th/U, La/Sc, and Th/Sc proxies, the Au-bearing stream sediments are sourced from felsic igneous rocks. These indications are corroborated by the correlation matrix assessment. However, Au is not sourced from the same source rocks as the host sediments due probably, to a prior depositional mixing of the sediments before subsequent transportation to their current depositional environment. With R² (0.62), MAE (0.6035), MSE (0.6546), and RMSE (0.8091) for LR, R² (1.0), MAE (0.7500), MSE (1.6273), and RMSE (1.2752) for DT, and R² (1.0), MAE (2.6608), MSE (12.7840), and RMSE (3.5755), for PR. The LR model performs better in predicting the Au occurrence in the area.