Application of Machine Learning for Prediction of Early Seven-Day Strength of Concrete in Factories

Abstract

In this study, several machine learning techniques were applied to predict the compressive strength of concrete on the 7th day. For the current research, samples were collected from 60 different concrete mixer plants in the Eastern province of Saudi Arabia. A total of 12 to 15 sets of samples were collected from each plant. Each set consisted of 6 cylinders; therefore, a database was established from tests on more than 567 samples. The data was used to validate and train three machine learning (ML) models: Credal Decision Trees (CDT), Multi-Layer Perceptron (MLP) and Naïve Bayes Classifiers. Modelling related to ML (specifically, CDT, MLP and Naïve Bayes) is used in the current study to see the possibility of predicting the compressive strength of concrete at 7 days. The Naïve Bayes performed the best with respect to prediction, Kappa stat and time consumption. The results show that Naive Bayes had the higher Kappa stat compared to CDT and MLP, indicating that NB has the best match between the trained and tested data. In addition, NB had the highest F-Measure score, which shows high precision. The results also show that NB has the highest Correctly Classified Instance among the algorithms tested, with respect to the time needed to build the model.