Strength and microstructural improvement of silica fume-modified coconut shell aggregate concrete using machine learning analysis

This paper analyses the mechanical behaviour and microstructure of Coconut Shell Aggregate Concrete (CSAC) with Silica Fume (SF) as an additional cementitious material. The binder was also partially replaced with SF at 5, 10 and 15 percent and 10 percent of the coarse aggregate replaced with coconut shells (CS). Out of all the mixtures, the one with 10% SF and 10% CS had the highest level of compressive, split tensile strength with an improvement of approximately 15% and flexural strength approximately 5% compared to the reference mix. This performance improvement can mainly be linked to the pozzolanic activity and filler effect of SF that led to the enhanced packing of the particles, increased the density of the matrix, and improved microstructural compactness. X-Ray Diffraction (XRD) was used to establish the greater proportion of amorphous silica, and the generation of secondary calcium silicate hydrate (C-S-H), and provided a qualitative understanding of the densification of the matrix after the introduction of silica fume by SEM analysis. Besides this, various machine learning (ML) algorithms were created; among them, Linear Regression, Support Vector Regression (SVR), Decision Tree, Random Forest, and XGBoost were used to predict the mechanical performance of the mixes, depending on the mixes and curing time. Out of all these models, XGBoost had the highest accuracy in the predicted values (R 2 = 0.96, RMSE < 1.5 MPa), which demonstrated the exploratory potential of artificial intelligence for predictive modelling of sustainable lightweight concrete mixtures.