Binary identification of Ma5 and He8 members in the Ordos basin using small-sample geochemical data and machine...

To address the challenges of limited sample size, drilling cuttings and the difficulty in horizon division, this study aims to explore a high-precision horizon identification method based on machine learning. First, systematic preprocessing was conducted on the major and trace element data of rocks, including data cleaning, data standardization, and feature engineering. Subsequently, the Random Forest (RF) algorithm was used for feature selection to screen key geochemical indicators, and two supervised learning models-XGBoost and Support Vector Machine (SVM)-were employed to construct lithological identification classifiers, whose performance was compared through fivefold cross-validation. The results demonstrate that the XGBoost model achieves optimal performance in key metrics such as accuracy and F1-score, with an overall identification accuracy of 0.9667. Combined with feature importance analysis, key controlling elemental indicators for horizon differentiation (e.g., Br and SiO₂) were identified. This study validates the effectiveness of machine learning in horizon identification using small-sample geochemical data, providing a novel data-driven perspective for fine-grained stratigraphic division.