Target-scale prospectivity modeling for gold mineralization within the Rajapalot Au-Co project area in northern Fennoscandian Shield, Finland. Part 1: Application of knowledge-driven- and machine learning-based-hybrid- expert systems for exploration targeting and addressing model-based uncertainties

This two-part paper reports results of data integration using different machine learning methods for target-scale prospectivity modeling of gold mineralization. The study area is the Rajapalot project area in the Northern Fennoscandian Shield in Finland. In this Part 1 paper, prospectivity modeling was implemented using knowledge-driven- and hybrid- expert systems. Prospectivity analysis-related systemic and model uncertainties were addressed using statistical and machine-learning based methods. The knowledge-driven Fuzzy Inference Systems (FISs) were optimized using Monte Carlo Simulations (MCS) to address and quantify mathematical model-related uncertainties and consequently present the prospectivity values at varying confidence levels. In the FIS results, as the confidence level increases, the high-prospectivity zones mapped by the models are spatially reduced and the mineralization trend becomes progressively well-defined. The resulting prospectivity map has 80% capture efficiency. The supplementary prospectivity map reporting prospectivity values at 90% confidence level has enhanced area-reduction capability. Machine-learning based Adaptive Neuro Fuzzy Inference System (ANFIS) was optimized under the neuro-fuzzy environment. This hybrid ANFIS-based result has 88% capture efficiency. Three exploration targets identified by the prospectivity maps together comprise about 1.5 km2 of the ∼20 km2 study area (i.e., 7.5% of the study area). This indicates considerable search-space reduction by the modeling results for follow-up ground exploration. Hence demonstrating that the prospectivity mapping methods, which are primarily ‘predictive tools’ at the regional scale, can be harnessed to serve as an aid to detection technologies at the target scale. Implementation of knowledge-driven and hybrid approaches deliver results that can (i) incorporate human cognition into prospectivity modeling, (ii) reduce bias of conceptual model, and (iii) are not overfitted to the training dataset. Statistical methods such as hypothesis testing and MCS, and machine learning methods such as ANFIS are therefore collectively beneficial for addressing both systemic and fuzzy systems-related modeling uncertainties.