Abstract:
Ore grade estimation is one of the most difficult problems when mining offshore deposits. In this research, a platinum deposit near Goodnews Bay, South-West of Alaska, was estimated with emerging techniques such as conditional simulation and support vector machines (SVM). Results of the estimation are presented and compared with a traditional estimation technique, such as the Inverse Distance Squared method. The area was divided in three clusters, based on the K-means method and geographical features. Also, Genetic Algorithm was used for appropriate data division. SVM parameters were optimized prior to the ore grade calculations. All estimations produced similar results for various cut-off grades, being the highest tonnages of platinum between 400 and 200 mg/m3. Reliability for conditional simulation was measured selecting blocks over 90% of probability for each cut-off value. SVM performance was evaluated using the Mean Square Error (MSE), the Mean Absolute Error (MAE) and the Correlation Coefficient Squared (R2). Using Pearson's correlation coefficient, SVM results presented a confidence of 95% in cluster 01, and 99% in clusters 02 and 03. Support vector machines seem to be an adequate tool for ore grade prediction, and it has a potential for its use in more complex geological and mining problems.
