Optimizing SVM Parameters for TBM Penetration Rate Prediction: Comparative Impact of Different Techniques on Model Performance

In this study, a Support Vector Machine (SVM)-based model was developed to predict the Rate of Penetration (ROP) during tunnel excavation. The model demonstrated high accu-racy and stability on both training and testing datasets, with performance metrics indica-ting its reliability (R² = 0.9583–0.9664, NSE = 0.9164–0.9292, MAE = 0.095–0.0968). To en-hance predictive performance, three systematic hyperparameter optimization strate-gies—Grid Search, Random Search, and Bayesian Optimization—were employed. Notably, Bayesian Optimization achieved high accuracy and computational efficiency with fewer evaluations, leveraging a probabilistic search framework and Gaussian Process-based modeling. Unlike previous studies in the literature, the dataset and input parameters used in this work exhibit greater diversity, and the effect of hyperparameter optimization on model performance was analyzed in detail. The results demonstrate that careful hyperparameter tuning can ensure strong generalization even under limited data conditions. This study provides significant methodological contributions to TBM performance prediction and ge-otechnical engineering applications.