Predictive modeling for sustainable high-performance concrete from industrial wastes: A comparison and optimization of models using ensemble learners. (10th April 2021)

• Record Type: Journal Article
• Authors: Farooq, Furqan; Ahmed, Wisal; Akbar, Arslan; Aslam, Fahid; Alyousef, Rayed
• Is Part Of: Journal of cleaner production. Volume 292(2021)
• Publication Date: 2021-04-10
• Subjects: Concrete strength -- Machine learning -- Prediction -- Artificial intelligence -- Supervised machine learning -- Ensemble model in concrete -- Waste materials; Factory and trade waste -- Management -- Periodicals; Manufactures -- Environmental aspects -- Periodicals; Déchets industriels -- Gestion -- Périodiques; Usines -- Aspect de l'environnement -- Périodiques; 628.5
• DOI: 10.1016/j.jclepro.2021.126032
• Languages: English
• ISSNs: 0959-6526

Abstract

Abstract: The cementitious matrix of high-performance concrete (HPC) is highly complex, and ambiguity exists with its mix design. Compressive strength can vary with the composition and proportion of constituent material used. To predict the strength of such a complex matrix the use of robust and efficient machine learning approaches has become indispensable. This study uses machine intelligence algorithms with individual learners and ensemble learners (bagging, boosting) to predict the strength of (HPC) prepared with waste materials. This is done by employing Anaconda (Python). Ensemble learner bagging, adaptive boosting algorithm, and random forest as modified bagging algorithm are employed to construct strong ensemble learner by incorporating weak learner. The ensemble learners are used on individual learners or weak learners including support vector machine and decision tree through regression and multilayer perceptron neural network. The data consists of 1030 data samples in which eight parameters namely cement, water, sand, gravels, superplasticizer, concrete age, fly ash and granulated blast furnace slag were chosen to predict the output. Twenty bagging and boosting sub-models are trained on data and optimization was done to give maximum R 2 . The test data is also validated by means of K-Fold cross-validation using R 2, MAE, and RMSE. Moreover, evaluation of ensemble models with individual one is also checked by statistical model performance index (e.g., MAE, MSE, RMSE, and RMLSE). The result suggested that the individual model response is enhanced by using the bagging and boosting learners. Overall, random forest and decision tree with bagging give the robust performance of the models with R 2 = 0.92 with the least errors. On average, the ensemble model in machine learning would enhance the performance of the model.