Hybrid Deep Learning Approach for Accurate Prediction of Flowability in Ultra-High-Performance Concrete
Abstract
By implementing several machine learning (ML), deep learning (DL), and hybrid deep learning models, the research methodology included a systematic approach, which included data separation, exploratory data analysis (EDA), artificial neural networks (ANN), K-Nearest neighbors (kNN), convolutional neural networks (CNN), long short-term memory (LSTM), Gated recurrent units (GRU), and convolutional neural network long short-term memory/gated recurrent units hybrid models. Also, the mean absolute error (MAE), R-squared (R2), and Root Mean Square Error (RMSE) were utilized to evaluate these models. Our results demonstrate that hybrid deep learning models, specifically the CNN-GRU configuration, achieve better performance in predicting ultra-high-performance concrete (UHPC) flowability compared to individual Deep Learning models and traditional Machine Learning approaches. The CNN-GRU model exhibited the best predictive accuracy with a RMSE of 1.360066 and MAE of 1.036573. Additionally, feature selection techniques enhanced the performance of certain models, with the feature-selected random forest model showing notable improvements in accuracy, achieving an RMSE of 1.032841 and MAE of 0.767066. Infrastructure durability and building processes can be improved with higher Ultra-High-Performance Concrete flowability prediction, which improves the effectiveness of various operations of the UHPC mixture design and benefits the application.