Optimizing Prediction Accuracy of Concrete Mixture Behavior Using Hybrid K-means Clustering and Ensemble Machine Learning

Abstract

Concrete is the most used material in infrastructure development, especially in a developing country. The concrete used in project must not only satisfy the desired concrete strength, but also the workability. Additionally, due to different conditions in construction projects, the requirement for workability varies. Workability can be measured using several methods. Previously, traditional trial-and-error of concrete mix design were used to achieve desired slump and flow test value. However, the experiment is often inexpensive, and the obtained results may not be sufficiently accurate. Recently, the potential of the AI method has been gaining increased attention as the new and promising alternative method to predict slump and flow tests, based on historical data. Thus, this study develops an effective hybrid AI-based method to predict slump and flow tests from the given concrete mixture dataset. A total of 103 historical data are used. At the beginning, the samples are separated into two groups using k-means clustering. Each cluster is modelled using the ensemble of six prediction methods, which are REG, CART, GENLIN, CHAID, ANN and SVM. The obtained results show that our proposed method can build the prediction method with a high accuracy, measured by several performance indicators.