Elemental analysis of the mining exploration data is important for many geochemical applications. The objective of this work is to conduct a comparison analysis of machine learning algorithms in predicting recovery of copper (Cu) from the Kazakhstan field. The data of cores was measured using the portable X-ray fluorescence (XRF) and the laboratory devices. We focused on the supervised machine learning algorithms such as K nearest neighbors (kNN), Decision Trees, Random Forest, XGBoost and LightGBM. The cross-validation result of these models shows that the Random Forest method can be used in the accurate prediction of the Cu recovery based on traditional laboratory tests. The examination of the algorithms is performed by metrics such as root mean square error, R2, and MAPE. In addition, the evaluation metrics of XGBoost and LightGBM are close to result of the Random Forest. In addition, the results are consistent with outcome of the SHapley Additive exPlanations (SHAP) framework.
Geochemical exploration using modern tools is essential for a wide range of applications such as mineral exploration, regional geological mapping, agriculture and forestry, baseline environmental monitoring, environmental restoration, and general land use management.
One of tools is X-ray fluorescence (XRF), which is a common method for analyzing geological samples (Carr et al., 2016). The handheld XRF provides fast measurements during the drilling process. These devices measure the fluorescent X-ray radiation that a sample emits when illuminated by a powerful X-ray source. Received XRF data used to determine geological characteristics and optimize well placement of samples.
Exploration of the geochemical data traditionally includes process of the sample preparation, logistical transportation of samples to the laboratory, and analysis with other available sources, etc. To reduce the exploration time, the combination of data from portable XRF and machine learning model can be useful in the identification of Cu recovery from the geological samples.
