Inferring Cenozoic Cover Thickness and Bedrock Sedimentary Material in 3D From Geophysical Data Using Machine Learning Algorithms: A Case Study in the Lockington Region, Victoria, Australia

Abstract This study presents a supervised machine learning approach to constructing a 3D geological model for the Lockington area in Victoria, Australia, by integrating borehole observations, geophysical surveys (magnetic, gravity, and radiometric), and elevation data. Two applications of machine learning are developed: (a) Cover‐bedrock models inferred the Cenozoic cover thickness, and (b) bedrock lithology models inferred the bedrock material as either pelitic or psammitic. The methodology involved data preprocessing through filtering and classification, as well as iterative model development using well‐established machine learning algorithms, such as Support Vector Machines and K‐Nearest Neighbors. We evaluated the importance of contributions from variously filtered geophysical survey maps using chi‐square scores, retaining the most influential features for model optimization. The cover‐bedrock models achieved an accuracy of 97.7%, while the bedrock lithology models achieved an accuracy of 92.6%, showing the approach's efficacy in capturing complex geological patterns and relationships. The final 3D models delineate the orientation of a domed anticlinal structure beneath the Cenozoic cover. These structures are consistent with existing geological interpretations of the area, as well as resistivity inversion pseudo‐sections, perpendicular to the D1 axial fold planes, that were not used as input to our models.