Predicting the prospect of rare earth elements and yttrium recovery from coal using machine learning

Coal and its coal by-products are important potential sources of rare earth elements and yttrium (REY), and the REY prospect in coal can be assessed by the outlook coefficient. However, REY data are lacking for some coal samples, and direct methods to determine REY content in coal are expensive and time-consuming, making it difficult to evaluate the REY prospect in these coals. To address this issue, this study integrates four popular machine learning algorithms (random forest, k-nearest neighbors, support vector machine, and XGBoost) with four feature selection methods (chi-squared test, F-test, mutual information, and XGBoost feature importance ranking). A total of 124 coal samples in China, collected from nine coalfields, were employed to train and optimize predictive models for assessing REY prospect in coal. The results show that the support vector machine, together with the chi-squared test feature selection, achieved the highest performance, with a classification accuracy of 0.996. In addition, this optimal model was further used to predict REY prospect in coal from the Heshan Coalfield in Guangxi Province in southern China and the Datanhao mine in Inner Mongolia in northern China. These predictions are consistent with the results obtained via direct analytical methods.