An Automatic Classification Method for Igneous Rock Fractures Based on an Interpretable Ensemble Machine Learning Model

Fractures play a pivotal role in controlling the reservoir quality and fluid flow capacity of buried hill hydrocarbon systems in igneous rocks. Accurate identification and evaluation of such fractures are therefore critical for efficient hydrocarbon exploration and development. In the Huizhou Oilfield, the spatial distribution of fractures within the igneous buried hills is highly stochastic, making reliable fracture prediction particularly challenging. To address this, a multidisciplinary approach involving core observations, thin section analysis, and formation micro-imager (FMI) logging was employed. Based on a comprehensive analysis of geological and logging data, fracture types in the study area were systematically characterized. To enhance classification accuracy, we propose a fracture-type discrimination model based on a Stacking ensemble learning algorithm. In this framework, five machine learning algorithms—KNN, Random Forest, XGBoost, SVM, and LightGBM—were first implemented as base learners. Five-fold cross-validation and PSO were applied to determine the optimal parameter set. Model performance was evaluated using four standard metrics: precision, recall, F1-score, and AUC. The final ensemble was constructed with Random Forest, XGBoost, SVM, and KNN as first-layer learners, and LightGBM as the meta-learner. Results show that the Stacking model outperforms all individual classifiers across all evaluation metrics, achieving an identification accuracy of 88.2%. This approach provides a robust solution for fracture detection and classification, particularly in scenarios where imaging log data are unavailable.