An ensemble approach to improve BPNN model precision for predicting compressive strength of high-performance concrete

Abstract

The use of machine learning and deep learning models to predict the properties of concrete is an emerging field to investigate. The effectiveness of these model's predictions depends on several parameters such as data set quality and amount, model type, input and output correlations, and many more. The better the model accuracy, the broader its reach in real-world applications. Therefore, this work intends to develop a distinctive deep learning model based on an ensemble technique to forecast the compressive strength of high-performance concrete; consequently, predictions are more accurate and trustworthy. The Artificial Neural Network (ANN) model is applied in the study to construct an ensemble model. Furthermore, Particle Swarm Optimization (PSO) methodology is utilized to optimize the hyperparameters of the ensemble model. The ideal number of successively coupled ANN models and learning rate obtained through PSO are 3755 and 0.0951, respectively. The feature significance research has also been undertaken to capture the effect of components on the compressive strength of concrete. The present ensemble technique has increased the model accuracy by 41.04 % compared to the single ANN model. The Root Mean Squared Error (RMSE) value of the proposed ensemble model is 4.009, and the coefficient of determination (R2) value is determined to be 0.93. The feature importance study indicated that concrete age, cement content, and water content are the key factors that influence concrete strength. The suggested ensemble model may be applied to decline the laboratory dependence for finding concrete properties.