Explainable Prediction Model of Logging Lithology Classification Based on XGBoost and SHAP

Lithology identification is a key and difficult problem in geological interpretation of reservoirs. To accurately predict the rock types in the target area of oil and gas exploration and provide correct reservoir lithology, an explainable model of lithology classification combining XGBoost and SHAP has been proposed. The model can achieve both good classification and explanation ability. Firstly, the lithologic classification model is established by the limit gradient lifting (XGBoost) algorithm, and the K-fold cross-validation (K-fold CV) and grid search algorithm are used to find the optimal hyperparameter combination of the model. Then, the predicted value of the optimized XGBoost model is compared with the random forest algorithm and support vector machine algorithm by using different evaluation indexes, and the overall and local explanation of the predicted result of the XGBoost model is given by using the SHAP method. The experimental results show that the lithologic identification model established by the XGBoost method optimized by hyperparameter adjustment has achieved good effects in the actual data test, and the SHAP method can effectively explain the output of the XGBoost model.