Mineral mapping of a gold prospect using ordinary cokriging and support vector machine algorithm: case of the Tikondi gold permit (eastern Cameroon)

Abstract

This study applied geostatistical and machine learning models, namely ordinary cokriging (OCK) and the support machine vector (SVM) algorithm, for mineral mapping of a gold prospect at Tikondi (East, Cameroon). For this purpose, five hundred and fifty (550) soil samples were collected and analyzed for Au, Ag, Zn, Fe, Cu, Pb, As, Sb, W and Bi. OCK and SVM models were validated using numerical and graphical methods of validation. Results showed that the gold grade ranged from 1 to 2480 ppb, with an average value of 9.973 ppb. The principal component analysis (PCA) demonstrated that bismuth (Bi) has the strongest association with gold grades. For OCK, the histogram of errors indicated a solid assessment when the root mean square error (RMSE = 21.41), mean absolute error (MAE = 4.76) and correlation coefficient (R = 0.841) indicated that OCK is a decent model, but with certain values poorly predicted. The confusion matrix and ROC measurement indicated clearly that SVM was a robust and efficient predictor for prospect mapping.