Clustering of Areas Prone to Iron Mineralization in Esfordi Range based on a Hybrid Method of Knowledge- and Data-Driven Approaches

In this study, a hybrid approach is established for clustering the most favorable regions  in association with magnetite-apatite mineralization at the Esfordi district in the central Iran. An optimum number of clusters is derived from a data-driven methodology through a concentration-area (C-A) fractal model of a synthesized geospatial data set. According to the metallogenic characteristics of the sought deposits, nine evidential layers deriving from geological, geochemical, geophysical, and remote sensing data were extracted. Prediction-area curve (P-A) was used as a data-driven method to determine the weight and importance of those evidences; then an index overlay method integrated them into a single propsectivity map. The number of clusters significantly affects the mineral potential modeling results in clustering algorithms. To determine an optimum number of clusters, the C-A fractal curve of the overlaid map indicated the correct population within this district, and then used as the optimal number to run the unsupervised clustering algorithms. Assuming five clusters, three clustering algorithms, including K-means, fuzzy C-means, and self-organizing map (SOM), were used to identify and localize iron-bearing favourable areas. The K-means algorithm had the highest accuracy in identifying those potential areas, by which 8% of the whole area could predict 65% of known deposits in the main favorable region.