Journal
STRUCTURES
Volume 53, Issue -, Pages 119-131Publisher
ELSEVIER SCIENCE INC
DOI: 10.1016/j.istruc.2023.04.062
Keywords
Tree-based algorithm; Bayesian optimization; Fiber reinforced concrete; Impact load; Machine learning
Categories
Ask authors/readers for more resources
This study develops machine learning models using tree-based algorithms and ensembles to classify the local damage levels of FRC panels subjected to missile impact load. Six different algorithms, including Decision Tree, Random Forest, Bagging, AdaBoost, XGBoost, and CatBoost, were trained and evaluated based on a dataset of 176 experiments. Bayesian Optimization algorithm and k-fold cross validation were utilized to improve prediction accuracy. The results show that the proposed models can predict the local damage levels with acceptable accuracy. Ensemble methods outperform single estimator models, and Random Forest is recommended for imbalanced datasets.
This study aims to develop the machine learning models for classification of the local damage levels of FRC panels subjected to missile impact load using the tree-based algorithms and ensembles. Six different algorithms, including Decision Tree, Random Forest, Bagging, AdaBoost, XGBoost, and CatBoost were trained and evaluated based on a dataset collected from 176 experiments of FRC panels under missile impact, which consists of 15 input parameters of geometries, materials, and boundary conditions and one output parameter of local damage level of FRC panels. The Bayesian Optimization algorithm and k-fold cross validation were also utilized to achieve higher accuracy in prediction ability of the models. The obtained results showed that the proposed models can predict the local damage of FRC panels subjected to the missile impact load with acceptable accuracy. The models using ensemble methods have better performance than single estimator model in prediction and each model using ensemble method has its own strength and is suitable for different criteria when classifying. For imbalanced dataset, Random Forest can be chosen as the most suitable classification model for the dataset in this study.
Authors
I am an author on this paper
Click your name to claim this paper and add it to your profile.
Reviews
Recommended
No Data Available