An orientation survey for methodizing classification accuracy of Cu mineralization by hybrid methods of fractal, neural network, and support vector machine in Haftcheshmeh, NW Iran

摘要

The main objective of this study is to compare and find an optimal method (structured or unstructured) for determining lithogeochemical and alteration classes of Haftcheshmeh Cu-porphyry deposit located in NW of Iran. Initially, fractal model of concentration-area (C-A) was applied to Cu data followed by principal component analysis in which PC 2 pertinent to Cu mineralization, utilized in C-A model. Both methods had weak results probably due to insufficient elimination of syngenetic effect by this method. To overcome these drawbacks, other supplementary models were implemented commencing with fuzzy C means clustering (FCMC), then PCA was applied to the residuals of FCMC. On the other hand, Neural Network (NN) classifier was added to optimize classification. A total of three integration methods, PCANN, FCMCNN, and FCMCPCANN, were executed for geochemical and alteration classifications. Among them, FCMCNN had the least mean squared error (MSE) and compatible results with the reality of the area. Furthermore, the integrated models of Support Vector Machine (SVM) such as PCASVM, FCMCSVM, and FCMCPCASVM were attempted and compared with the results of NN integrations. The SVM results were unsatisfactory due to low classification accuracy, whereas the NNs methods had privileges for classification. Overall, comparative stepwise approach indicates FCMCNN as tangible optimized technique in eliminating the syngenetic effects causing considerable uncertainty reduction for classifying different geochemical classes enabling the proposed method more reliable for detailed exploration.