Evaluación del potencial estratégico de Uranio (U) y Torio (Th) mediante algoritmos de clasificación de machine learning (GMM y SC) y análisis litogeoquímico

Abstract

Abstract: The prospecting of Uranium (U) and Thorium (Th) constitutes a strategic component in the exploration of energy minerals, which has traditionally been addressed through geostatistical interpolation or supervised classification, with limited integration of lithogeochemically validated unsupervised learning algorithms. To bridge this methodological gap, this study evaluated the strategic potential of U and Th by integrating Ordinary Kriging with Gaussian Mixture Models (GMM) and Spectral Clustering (SC) algorithms, utilizing a dataset of 4000 surface lithogeochemical samples (Easting, Northing, Elevation, U, Th, and ten lithological units) from a region in Ecuador. Following a log-transformation of the elemental concentrations, exponential variograms were fitted (with ranges of 1314.9 m and 585.9 m) and interpolated over a grid consisting of 10,890 nodes. Subsequently, the area was classified into potential zones using GMM (BIC, k=6, post-hoc collapsed to three levels) and SC, and their spatial concordance was compared (ARI = 0.674). The lithogeochemical validation confirmed a significant association between lithology and mineral potential (Chi-square p < 0.001; ANOVA F = 314.60 and 108.02, p < 0.001). It is concluded that the integration of geostatistics and machine learning generated automated, statistically validated prospectivity maps, identifying the Jub unit as the primary favorable lithological control.