Mineral resource classification should consider multiple quantitative and qualitative criteria related to the geological and grade confidence, data quantity and quality, and the prospective mining method. Basing mineral resource classification on a
single quantitative criterion is seldom adequate, and often leads to suboptimal
results. However, integrating multiple different and sometimes divergent criteria is usually challenging and subject to inconsistency, that is, blocks with similar qualities may end up in different categories. To address this challenge, SRK has successfully implemented machine learning clustering algorithms to develop more comprehensive mineral resource classification schemes. Applying machine learning terminology, these algorithms can be applied in a semi-supervised and unsupervised way, with results always subject to review and editing, if necessary. The result is a consistent classification into mineral resource categories that is comparable to the result of a classification done by a conventional approach, but fully consistent and generated in a short timeframe. The workflow can be easily reproducible; therefore, the result can be easily audited.
The semi-supervised mineral resource classification approach consists of an unsupervised stage followed by a supervised stage. In the first stage the block model is divided into different randomly selected subsets. For each subset a clustering algorithm creates groups of blocks according to how similar their corresponding estimation metrics are. These metrics may include multiple estimation results and parameters, such as the average distance to informing samples and/ or drill holes, estimation variances, number of kriging passes, and many others. In the second stage the resulting clusters are used as reference data for classifying the whole block model. By repeating this stage for multiple subsets, it is possible to obtain each block’s probability of belonging to each of the mineral resource categories. Finally, a smoothing algorithm is applied to define the final boundaries between mineral resource categories.
For unsupervised classification, a score is given to multiple qualitative and quantitative criteria. Depending on each criterion, a higher score implies higher confidence in the estimation, in the quality of the information, or in the prospects for eventual economic extraction. A weighted average score is calculated with weights assigned to each criterion at the discretion of the Competent Person. The average score is rescaled by the level of geological confidence communicated through the geological model. An unsupervised clustering algorithm creates three spatially correlated clusters and support vectors are used to automatically smooth the boundaries between clusters. Results are validated by the classification criteria statistics and reviewed and edited by the Competent Person in a series of cross sections to produce the final mineral resource category boundaries.
Watch a video of this topic here.
