Abstract
This article represents part of a broader research project aimed at developing predictive technologies for identifying prospective mineralized zones based on the analysis of data from an integrated subsurface use platform. The study presents a predictive modeling framework for lithium mineralization within the Kalba–Narym metallogenic zone using machine learning and geostatistical methods. The scientific novelty of the research lies in the integration of geochemical, radiometric, and geophysical data extracted from a cloud-based geospatial platform into a unified mineral prospectivity prediction system. Random Forest (RF), Gaussian Process Regression (GPR), and Empirical Bayesian Kriging (EBK) were applied to predict lithium concentration and analyze spatial patterns. The input data included geochemical indicators, radiometric data, magnetic anomalies, and gravity data. Prior to modeling, all datasets were harmonized into a unified spatial and numerical format. The calculated anisotropy ratio (AR) values revealed the presence of direction-dependent spatial continuity and directional asymmetry within the studied fields. At the same time, the overall similarity of variogram shapes across different directions indicates coherent and structured spatial organization rather than completely random variability. The RF model demonstrated greater effectiveness in identifying localized lithium enrichment anomalies, whereas EBK and GPR better represented regional spatial trends and continuity. The resulting prospectivity maps show spatial correspondence between elevated lithium concentrations and gravity, magnetic, and radiometric anomalies. Five prospective lithium mineralization zones were identified within the study area: East Kalba, Central Kalba, Yeser, Proletarsky, and Kovalevsky. The obtained results confirm the effectiveness of integrating machine learning and geostatistical approaches for rare-metal prospectivity mapping and may support future mineral exploration planning.