MACHINE LEARNING APPLICATION IN PREDICTIVE MINERAL MAPPING OF SOUTHWESTERN UGANDA: LEVERAGING AIRBORNE MAGNETIC, RADIOMETRIC, AND ELECTROMAGNETIC DATA

This research delves into integrating machine learning with traditional geological methods to overcome mineral exploration challenges, particularly in ore and rare earth element exploration within the Mesoproterozoic complex of southwestern Uganda. While crucial, traditional geological mapping faces limitations due to site accessibility, basement exposure, financial constraints, weather conditions, and private sector limitations. Our study employs advanced deep-learning techniques alongside traditional geophysical data analysis from airborne magnetic, radiometric, and electromagnetic surveys to address these challenges. This approach aims to enhance the precision of mineral mapping predictions. The methodology includes the application of various filters and derivatives, including reducing to the magnetic pole (RTP), first vertical (1VD) and horizontal derivatives (HDR) in the x and y axes, tilt derivatives (TDR), and applying the analytical signal (AS) to the magnetic data. Central to our research is using a supervised deep neural network predictive targeting (SDNN-PT) methodology combined with traditional geophysical mapping. This approach aims to identify potential ore mineralization zones and elucidate the controls of ore localization within the complex geological framework of the region through traditional methods and then predictive machine learning methods. Our findings indicate two areas of significant interest where the pegmatites proximal to granite exhibit pronounced magnetic lineaments derived from the first vertical and tilt derivatives, indicating potential ore mineralization in linear structures such as faults, shear zones, or boundaries between rock types. These areas serve as vital training zones for our deep-learning neural networks, informing predictive mapping efforts in other areas of interest and further refining interpretations from traditional methods. This study contributes to geoscientific research and offers potential economic benefits through improved exploration strategies in Uganda's mineral-rich regions by bridging traditional geologic methods with machine learning.