Abstract
The rapid development in construction industry, induce a large amounts of concrete data that are usually measured and analyzed everyday naming that concrete is the second usable material on earth. Concrete is made from numerous ingredients that have huge variability either at the design stage or at the testing stage. The main goal of this paper is to quantify the anomalies and outliers during the design phase of concrete mixtures. Concrete mixtures have various percentages of ingredients such as cement, slag, fly ash, water, superplasticizer, fine and coarse aggregates. Machine learning and data mining is considered a very thriving topic in many research fields and its implementation in the construction industry still limited. Concrete community is in need for such a tool to produce an efficient way to efficiently design concrete mixtures. Outliers could occur during the evaluation of samples' measurements that might include human or system errors. The Local Outlier Factor (LOF) algorithm is the most common method used to determine outliers, however, the LOF has some challenges. In this paper, an anomaly-based outlier detection algorithm called Isolation Forest based on a Sliding window for the Local Outlier Factor (IFS-LOF) algorithm, is proposed to solve the limitations of the LOF in evaluating 1030 concrete mixtures. The proposed algorithm works without any previous knowledge of data distribution and executes the process within limited memory and with minimal computational effort. The evaluation of results proved that the IFS-LOF algorithm is more efficient in detecting the sequence of outliers and provided more efficient accuracy that other state of the art LOF algorithms. (C) 2020 Elsevier Ltd. All rights reserved.