Machine learning model optimization with optuna for accurate prediction of strength and crack behavior in prestressed concrete beams

Prestressed concrete beams are widely used in bridge and building structures, and their performance is directly related to the overall safety and durability. To predict the performance of prestressed concrete beams, machine learning (ML) has been widely applied. However, the selection of model hyperparameters remains a significant challenge in achieving efficient and accurate predictions. To address this issue, this paper investigates the application of Optuna for hyperparameter optimization to four machine learning models: XGBoost, Decision Trees (DT), Random Forests (RF), and LightGBM (LGBM). This study demonstrates that adopting this optimization method significantly improves the efficiency of the prediction process and successfully identifies the optimal hyperparameter combinations for each model. The performance of the final models was validated using a set of performance evaluation metrics. The results show that the LGBM model achieved an R2 value exceeding 0.98 for strength prediction and over 0.8 for crack resistance prediction. Therefore, hyperparameter tuning using Optuna not only significantly improves the prediction accuracy of the models but also effectively reduces computational and time costs.