Prediction of displacement of composite slab with profiled steel deck using artificial neural network

Traditional assessment of composite slab is dependent on laboratory testing, a process that incurs significant expenses and consumes a considerable amount of time. The current research introduces a novel approach by leveraging machine learning to accurately predict the maximum displacement of composite slabs with profiled decks, thereby offering a simpler and more efficient alternative to physical testing. Artificial neural networks (ANNs) utilized specifically to exploit feedforward networks optimised by Bayesian regularisation (BR) to generate the present models to predict displacement of composite slabs with profiled steel deck. The models were trained and evaluated on a dataset consisting of 10 independent variables that affect slab displacement, as well as one dependent variable. This dataset was created by combining 210 training datasets and 40 testing datasets from prior research and synthetic data generated using generative adversarial networks (GANs). A thorough sensitivity analysis was used for verification, thereby highlighting the significance of the chosen parameters. Five ANN models were systematically built, with varied levels of complexity ranging from ten to forty neurons and assessed for their predictive skills using error analysis, regression parameter testing, and performance plots. The results of the current research reveal that all models achieved a high level of accuracy, with a coefficient of determination more than 98% and prediction errors below 10%. Blind tests conducted on randomly selected experimental datasets provided further validation for the models, emphasising the effectiveness of ANNs with twenty neurons in accurately forecasting the displacements of composite slabs with a negligible absolute and relative error.