Estimation of Decision Boundaries for Critical Zone Classification in a Polymetallic Tailings Dam Using Machine Learning

The objective of this study was to evaluate the performance of three machine learning models for classifying and delineating critical contamination zones in a polymetallic tailings pond. Four hundred samples (water and soil) were analyzed using physicochemical variables (pH, electrical conductivity (EC), lead (Pb), and copper (Cu)). The methodology implemented Random Forest (RF), Support Vector Machine (SVM), and K-Nearest Neighbors (KNN), evaluated through 10-fold cross-validation, reporting the mean and standard deviation. The results showed that complexity is matrix-dependent: water data exhibited linear separability, allowing for perfect classification (1.0 ± 0.0), while soil data showed non-linear overlap. In this complex scenario, RF emerged as the most robust model, achieving an accuracy of 0.980 ± 0.033 and an F1-score of 0.989 ± 0.019, surpassing the stability of SVM and KNN. It is concluded that RF is the most effective tool to minimize the risk of false negatives in spatial delimitation, guaranteeing accurate environmental remediation.