Comparing Classification of Concrete Flexural Strength Between Binary Relevance and Classifier Chains Algorithms

Abstract

Testing the flexural strength is a crucial issue for evaluating structural performance in impractical on-site conditions. This limitation requires more efficient methods to achieve concrete quality classification. This study aims to develop a flexural strength quality classification model using machine learning-based multi-label classification approaches, specifically Binary Relevance (BR) and Classifier Chains (CC) algorithms. A synthetic dataset representing the characteristics of concrete mixtures was used to train the classification models into three quality categories: Good, Fair, and Poor. The modeling process involved data preprocessing, label assignment, model construction using the Random Forest algorithm, and performance evaluation using metrics such as Hamming Loss, Subset Accuracy, F1-Score, and Jaccard Similarity. Experimental results show that the CC algorithm outperforms Binary Relevance across all evaluation metrics, achieving a Subset Accuracy of 74% and an F1-Score of 0.81. These findings demonstrate that the CC approach effectively captures label dependencies, making it a promising solution for more efficient and accurate concrete quality assessment in construction practices.