Deep Hybrid Neural Network Models for Estimating Compressive Strength in Eco-Friendly Materials

The cement industry's significant carbon dioxide emissions considerably impact greenhouse gas levels, establishing it as a major polluter. Moreover, the extraction of natural resources for concrete manufacture has been excessive due to rapid population growth and escalating construction demands. This depletion has created an urgent need for environmentally benign alternatives that do not compromise compressive strength. The paper examines four data transformation methods: statistical normalisation, min-max normalisation, non-linear transformation, and whitening transformation. The Autofeat approach is employed for automatic feature engineering and selection to enhance data preprocessing. The application of inter- and intra-periodic convolutional blocks facilitates the extraction of superficial features from sequences of varying dimensions utilising a CNN model with a DSDB architecture. To enhance the extraction of profound sequence information, a transformer module using GWT captures comprehensive temporal dependencies. The proposed approach facilitates the prediction of the compressive strength of eco-friendly materials with an accuracy of 97.46%. These results illustrate that advanced data transformation and feature extraction enhance the prediction of material performance, highlighting the potential of deep learning to optimise sustainable construction materials, thereby facilitating more efficient and environmentally sustainable building solutions.