Using Machine Learning for Automatic 3D Anomaly (zone) Identification
https://minervaintelligence.com/driver-primer-3-creating-zones/
In our last blog post, we showed how DRIVER can be used to quickly and easily create accurate block models of all a dataset’s attributes.
In this post, we will discuss how DRIVER can automatically — and virtually instantaneously — identify anomalous zones in your deposit.
In mineral deposit modelling, continuous numeric variables that delineate 3D volumes or “zones” of importance (e.g., mineralisation, alteration, lithology) are often defined manually using a cut-off, or threshold-based, grade shell. For example, a 3D volume (zone) of high gold (Au) concentration in this Saddle North dataset may be visualised by extracting all blocks that have a concentration > 1 ppm:
This method works well when dealing with only a few variables, but it becomes tedious and error prone when analysing multiple attributes, especially if the attributes and their concentrations of importance are not well understood. In other words, the average geologist knows a great deal about certain concentration thresholds for gold (Au) and copper (Cu) in a porphyry deposit, but what range of concentrations should be chosen to create 3D shapes representative of anomalous sulphur (s), arsenic (As) or zinc (Zn)?
Fortunately, this is a problem that can be solved using machine learning cluster analysis and spatial statistics.
DRIVER provides an algorithm that can automatically evaluate each block model and extract clusters of blocks that represent spatial and chemical anomalies. The DRIVER algorithm is flexible and can operate on a wide range of data, meaning that it can be set to run simultaneously on each of the estimation block models. To access it, right click the “Zones” section and open “Auto generate zones”. Each of the block models can be multi-selected for quick, parallel processing. It is often useful to set the “minimum concentration threshold” to the mean (average) value in the dataset, making sure that all anomalies identified are at least above average.
The zone objects are processed in DRIVER in less than a minute. The results are 3D wireframes that denote clusters of blocks that are anomalous relative to their local neighbourhood values. The results for Au for example, form a broadly similar shape to the manually constructed >1 ppm grade shell; however, the average concentration of the blocks inside of the machine-learning-generated Au zone is 1.03 ± 0.34 ppm. Its also important to remember that the machine-learning zone boundary is not set at a fixed threshold; rather, it has been placed as a function of local concentration change (gradient) in the data.
The key benefit to this method, however, is that DRIVER has generated useful 3D zones for every other element in the dataset, and the user is immediately able to start evaluating the broad geochemical patterns and features of the deposit. Bismuth (Bi), for example, shows a clear anti-correlation to Au. Copper (Cu) is strongly enriched in the main gold zone. Arsenic (As) may be an important penalty element in the ore. However, by using DRIVER, we can immediately see that the arsenic anomalies are not strongly spatially associated with the mineralised gold-copper zone.
This is the power of using DRIVER for multi-attribute geochemical analysis. By using machine learning and cloud processing, DRIVER can create 3D volumes of anomaly zones without the user needing to supply a single cut-off grade or having any specific knowledge about significant concentrations of each element. By facilitating rapid, quantitative geochemical analysis, DRIVER’s machine learning supported zone analysis tools provide enormous value for users, facilitating rapid, quantitative geochemical analysis that inform robust exploration and mining decisions.
In the next DRIVER Primer blog post we will be returning to DRIVER basics, covering data preparation, auditing, and uploading.