Optuna-Optimized GBDT Approach for Soil Stratification Analysis in Cone Penetration Testing

Traditional soil layer classification methods primarily rely on the combination of in-situ testing and expert evaluation. However, this approach is often associated with high costs and may introduce subjectivity during soil stratification. In recent years, with the rapid development of big data and artificial intelligence, machine learning has demonstrated unique advantages in geotechnical engineering. To overcome the limitations of conventional soil classification methods, this study proposes an Optuna hyperparameter-optimized Gradient Boosting Decision Tree (GBDT) algorithm model for soil layer classification prediction. Utilizing data obtained from Cone Penetration Testing (CPT), the dataset was divided into training and test sets, with four evaluation metrics employed to comprehensively assess prediction performance. Furthermore, the model was compared with several alternative algorithms including KNN, RF, MLP, XGBoost, LightGBM, and CatBoost. Results indicate that the Optuna-optimized GBDT model exhibits superior performance compared to other models, demonstrating higher prediction accuracy and stability. In practical engineering applications, the model achieved prediction accuracies exceeding 0.8, with some cases reaching over 0.9, indicating satisfactory predictive effectiveness. Therefore, the proposed Optuna-optimized GBDT algorithm provides an effective solution for soil layer classification and offers valuable references for further research in this field.