Concrete compressive strength prediction with machine learning

Resumo

Compressive strength is the main characteristic of concrete. The correct prediction of this parameter
means cost and time reduction. This work built predictive models for 6 different ages of concrete samples. A set
of 1030 samples from previous studies was used, with 9 variables. Another 6 variables were added to represent
the proportions of the main ingredients in each sample. The predictive models were developed in R language,
using the Parallel Random Forest algorithm and repeated cross-validation technique to optimize the parameters.
The results were compatible with other studies using the same data set. The most important model, 28 days old,
obtained a root mean square error (RMSE) of 4.717. The 3-day model obtained the best result, RMSE of 3.310.
The work showed that the compressive strength of concrete can be predicted. The choice of creating a model for
each age allowed to get compatible results with the available data at each age. It was a promising alternative since
good results were achieved by training with just one algorithm. This work facilitates exploration and new efforts
to predict the compressive strength of concrete, it can be used as a baseline to predict with different algorithms or
the combination of several.