Featured Application
The computational framework developed in this research—specifically, the multilayer perceptron artificial neural network architecture—presents an innovative workflow for implementation in geology and mining software such as RecMin. By leveraging database connectivity protocols, the Python-based neural network can dynamically read the coordinates of the block model and perform categorical classification. This integration could update the modeling approach by incorporating artificial intelligence algorithms, taking into account the findings from this research and considering boundary constraints to optimize the prediction of geological attributes.
Abstract
The delineation of mineralized zones within deposits is inherently associated with geological uncertainty, directly impacting resource-reserve estimation and eventual extraction. This study presents a comparative analysis of three geological modeling techniques, including Indicator Kriging (IK), Nearest Neighbor (NN), and Multilayer Perceptron Artificial Neural Networks (ANN-MLP). The dataset size comprises 2097 4-m composite samples derived from 51 exploration drillholes within a copper porphyry deposit. A 3D block model with a 10 m × 10 m × 10 m block size was developed with the purpose of providing a common framework for applying the modelling techniques and assigning the categorical attributes. The validation design integrated vertical Z-direction swath plots alongside multiclass calibration plots to benchmark spatial trend reproducibility and prediction confidence. Results demonstrate that while IK and NN preserve regional spatial trends and local variability, the data-driven ANN-MLP model enables the development of geological modeling for mineralized zones directly from spatial coordinates. Furthermore, the ANN-MLP model exhibits distinct operational features, displaying localized over-confidence and under-confidence calibration shifts, which highlights the opportunity to integrate boundary regularization constraints to optimize spatial prediction reliability.