Predictive Modelling of Nickel Potential with the Integration of Multisource Information Based on Random Forest

Organisation: Vale – University of Granada
Start and Estimated Duration: 02 – August – 2018, 24 Months

Summary: Mineral exploration activities require robust predictive models that result in the accurate mapping of the possibility that mineral deposits can be found at a specific location. Random forest (RF) is a powerful machine data-driven predictive technique that is unknown in potential mineral mapping. In this project, the performance of RF regression for the nickel deposits in the vale mine sites in Sudbury, Ontario, Canada is explored. The results of this project indicate that the use of RF for the integration of sizeable multisource data sets used in mineral exploration and for prediction of mineral deposit occurrences offers several advantages over existing methods. Key benefits of RF include the simplicity of parameter setting; an internal unbiased estimate of the prediction error; the ability to handle complex data of different statistical distributions, responding to nonlinear relationships between variables; the capability to use categorical predictors; and the capacity to determine variable importance. Variables that RF identified as most significant coincide with well- known geologic expectations. To validate and assess the effectiveness of the RF method, nickel prospectively maps are also prepared using the Logistic Regression (LR) method.