MACHINE LEARNING-BASED MODEL FOR PREDICTING CONCRETE COMPRESSIVE STRENGTH

Abstract

This study aims at applying a machine learning-based model to establish the relationship
between different input variables to the 28-day compressive strength of normal and High-Performance
Concrete (HPC). An Artificial Neural Network (ANN) model was trained, validated, and tested using a
comprehensive database consisted of 361 records gathered from the previously circulated source. Various
models with different learning algorithms and neuron numbers in the hidden layer were examined to attain the
best performance model. The examination results revealed that the ANN model using the “trainlm” learning
algorithm delivered the best prediction outcomes with the overall coefficient of determination (R2) of 0.9277.
The influence of input parameters on the output was also examined by performing the sensitivity analysis. It
was observed that the compressive strength of concrete at 28 days was more responsive to the changes in the
cement parameter (CM) and the amount of water (WT). In contrast, the 28-day concrete compressive strength
was found less sensitive to the variation of the fly ash (FL) parameter.