1. Optimizing Shear Capacity Prediction of Steel Beams with Machine Learning Techniques.
- Author
-
Elamary, Ahmed S., Sharaky, Ibrahim A., Alharthi, Yasir M., and Rashed, Amr E.
- Subjects
- *
PLATE girders , *MACHINE learning , *SHEARING force , *STEEL , *RESEARCH personnel , *FORECASTING - Abstract
The shear capacity of steel beams is important for safe and efficient design in construction. Existing approaches have not accurately predicted the ultimate shear resistance of plate girders, but machine learning (ML) has emerged as a new technique to address this challenge. The Lazy Predict Python library automates the model selection and hyperparameter tuning process. In a recent study, up to 42 ML models based on Lazy Predict were applied to two datasets from 100 test results of varying specimens conducted by previous researchers. The XGB regressor model proposed in this study provided the highest performance, achieving an adjusted R squared of 0.9262 and 0.9417 R squared for the first dataset and an adjusted R squared of 0.9513 and 0.9692 R squared for the second dataset outperforming the other ML techniques in scikit-learn library. The values obtained from each test data were multiplied by the specimen's estimated shear capacity using a theoretical technique, and these were compared with experimental results. The maximum shear force resulting from the proposed modified equation was calculated using two standards, EC3 and AISC proposal, to estimate the ultimate shear force. This study demonstrates that ML techniques can improve predictions of shear capacity in steel beams, which is crucial for safe and efficient construction design. The recommended model exhibited an appropriate level of accuracy in calculating the shear capacity of steel beams based on test data from earlier research and the current study. The datasets and codes utilized in this study can be freely accessed at https://github.com/amrrashed/Shear-Capacity-of-Steel-Beams-with-Flat-Webs-by-Using-Optimised-Regression-Learner-Techniques. [ABSTRACT FROM AUTHOR]
- Published
- 2024
- Full Text
- View/download PDF