The sandstone-type uranium deposits located within the Songliao Basin of China, which are noted for their significant reserves and low mining costs, have become a primary focus for uranium exploration in the country. Accurate detection of such anomalies is essential for the exploration and assessment of uranium resources. Traditional logging identification methods face challenges, including low accuracy, slow recognition speeds, and limited generalization capabilities. With advancements in technology, artificial intelligence has introduced a novel research paradigm for identifying uranium deposits.

This study, which concentrates on the sandstone-type uranium deposits in northern China’s Songliao Basin, employs two representative ensemble learning algorithm models, extreme gradient boosting (XGBoost) and random forest (RF), to facilitate the automatic identification of stratigraphic lithology and uranium-bearing layers. The performance outcomes of these models are compared with those from the K-nearest neighbor classification algorithm, the gradient boosting decision tree algorithm, the back propagation algorithm, and the support vector machine algorithm, which are recognized as typical machine learning algorithms. The prediction accuracy across all six models exceeded 91%, underscoring the efficacy of machine learning techniques in identifying lithologies associated with uranium deposits.

Among them, the XGBoost model demonstrated superior recognition performance with an accuracy rate of 98.54%, followed closely by the RF model at 98.20%. Both the XGBoost and RF models exhibited high accuracy rates in detecting uranium anomaly layers and mineralized zones, achieving accuracies of 98.81% and 98.22%, respectively.

To address issues related to imbalanced sample data, this study employed the synthetic minority oversampling technique, thereby enhancing both accuracy and comprehensiveness when identifying thin uranium-bearing layers. The optimization process grounded in ensemble algorithms provides a theoretical foundation as well as technical support for intelligent identification methodologies pertaining to sandstone-type uranium deposits.