Hyperparameter-Optimized machine learning approach for compressive strength prediction of metakaolin concrete reinforced with glass fiber

Abstract

This paper introduces an optimized machine learning framework for the prediction of the compressive strength of metakaolin concrete with glass fiber reinforcement based on the comparison between artificial neural network and deep neural network models. The innovative aspect of the proposed framework is the optimization of the hyperparameters of both models using Keras Tuner in order to obtain the best configuration of the networks and their training procedure. A set of 168 concrete mixtures was generated experimentally and was used to train and validate the models. The input features used were the quantity of cement, metakaolin, water, aggregate, superplasticizer, glass fiber, and curing time, and the compressive strength was used as the output. Predictive accuracy of the developed models was determined through mean square error, mean absolute error, and coefficient of determination. Both models exhibited high predictive accuracy and yielded identical outcomes. The developed artificial neural network model obtained mean square error of 11.29, mean absolute error of 2.70, and coefficient of determination of 0.892, whereas deep neural network model, obtained mean square error of 11.69, mean absolute error of 2.64, and coefficient of determination of 0.884. Even though the deep neural network yielded a slightly lower mean absolute error, the artificial neural network outperformed the deep neural network in terms of accuracy. This research framework can be regarded as an accurate and efficient alternative to traditional laboratory experiments for quick evaluation of compressive strength and subsequent design and optimization of sustainable concrete reinforced with metakaolin and glass fibers.

Graphical Abstract