Title:
Research on the Prediction of Compressive Strength of Marine Concrete Based on Support Vector Machine Algorithm and Bayesian Optimization (Prepublished)
Author(s):
Jingcheng Ye and Chunxiang Qian
Publication:
Materials Journal
Volume:
Issue:
Appears on pages(s):
Keywords:
bayesian optimization; compressive strength prediction; marine concrete; support vector machine
DOI:
10.14359/51751979
Date:
8/21/2026
Abstract:
The marine environment accelerates the deterioration of concrete’s mechanical properties, necessitating accurate prediction of compressive strength for service life and durability design. Using a marine concrete database, this study develops a Support Vector Machine (SVM) model to examine the impacts of data quality, dataset size, and hyperparameter optimization. Gray relational analysis identifies initial strength, silica fume content, SO₄²⁻ concentration, and exposure age as key factors. Noise simulations reveal that erroneous data significantly reduce predictive accuracy, underscoring the importance of data reliability. As the dataset expands from 500 to 10,000 samples, R² increases from 0.515 to 0.939, while RMSE and MAE decline markedly, though accuracy gains taper beyond 5,000 samples. Bayesian optimization delivers the best performance with an R² of 0.946 and an average relative error of 5.52%, demonstrating its superiority in enhancing model adaptability and prediction precision for marine concrete durability assessment.