Regularization Learning of Trace Element Contamination Stemmed from Tailings Dam-Break

An important practice in environmental risk management is assessing the consequences of heavy metal concentrations resulting from a mine dam tailing failure on soil, water, and trees. To appraise the extent of pollution, an effective classification is essential. In this study, trace element contamination is handled as a two-group classification problem and examined the performance of supervised regularization algorithms as spatial classifiers using imbalanced uncertain data. In addition to conventional shrinkage algorithms such as Ridge, the Lasso and Elastic-Net, the generalized t-statistic-based U-Lasso classifiers have been introduced and tested for mitigating such imbalances and adjusting weights for class distributions. The feature interpretation studies underlined that the most important indicator of the models is Zinc (Zn). The experimental studies revealed that the Ridge classifier (l2 penalty) outperforms the other models. Statistically, the U-Lasso models exhibited notable explanation capacity and their performances recorded close to the conventional shrinkage algorithms. The use of statistical learning-based classification approach to appraise geo-environmental contamination under the conditions of natural variability and spatial uncertainty provides useful meta-data and reliable classification models.