Multi-layer Perceptron for Prediction of Axial Strength of Fully FRP-Confined Circular Concrete

This study presents an effective utilisation of the Multi-Layer Perceptron (MLP) machine learning algorithm to predict the peak axial strength of axially loaded circular cross-section concrete columns confined by continuous fiber-reinforced polymer (FRP) jacket. In this study, to achieve the most reliable predictive model by using MLP method, a large experimental dataset consisted of 1517 samples obtained from available experimental axial compressive tests registered in the literature. For obtaining an efficient performance in MLP, the hyperparameter tuning is a vital process, which is based on random searching to classify the hyperparameter ranges. Once hyperparameter ranges are generated, the genetic algorithms are implemented to search for optimal hyperparameters that yield the best predictive performance. Furthermore, cross-validation with 10 folds is employed to avoid the overfitting and enhance the prediction performance of the MLP model. The performance of the developed MLP model is assessed by comparing with the existing design equations. The comparative assessment indicates that the implemented model with MLP provides superior performance in predicting the compressive strength of axially loaded FRP-confined circular shape cross-section concrete columns, compared to existing predictive design equations.