Improved cluster analysis of Werner solutions for geologic depth estimation using unsupervised machine learning

Werner deconvolution is a widely used method for computer -aided solution estimations using magnetic field data. Because of the inherent unpredictability of potential field data, the techniques frequently yield unpredictable results and may not always be able to predict the full extent of the geological structures depicted. The unsupervised clustering algorithm was then used to improve the detection of geological structures generated from magnetic field data. A synthetic magnetic model was created by combining two models that resembled dikes. To add a little more complexity to the model, sporadic noise was applied to the synthetic model. The synthetic model was subjected to Werner deconvolution to produce solutions. The developed solutions resulting from the synthetic data were applied to an unsupervised machine -learning algorithm. The algorithm was able to detect two clusters from the dataset and the centroids generated from the number of clusters formed indicated the depth of the two models, which was 5 m and 8 m respectively. The algorithm was then subjected to a real case dataset from the Banke complex, which is part of the Nigerian Younger Granite region. Three clusters were formed, having centroids of 536 m, 635 m, and 530 m. These depths were validated with previous surveys carried out in the area. This algorithm proved successful in determining the exact position in terms of the depth of the geologic bodies. This proposed approach is fully data -driven and is successful even when there is noise.