GA‐based multi-level association rule mining approach for defect analysis in the construction industry

• An effective method is proposed to extract useful knowledge for construction managers from previous defect cases.
• The traditional Apriori algorithm is improved with GA and a concept hierarchy.
• An automated rule pruning method was developed for association rule mining.
• A case study was conducted on the defect database of Chinese construction industry during 2000 to 2010.

In construction industry, work defects yield time and cost overruns of construction projects and also cause disputes between project participants during construction and operation phases. To date, there hasn't yet been an adequate analytical model to extract useful information from the database of construction defects. The information represented in the form of association rules could enhance quality management via defect prediction and causation analysis. This paper proposes a Genetic Algorithm (GA)-based approach that incorporates the concept hierarchy of construction defects to discover multi-level patterns of defects from the database of defects in the Chinese construction industry during 2000 to 2010. First, the domain knowledge of construction defect is incorporated into a concept hierarchy to adjust mining items at different levels according to the data sparseness and the interestingness of a rule. Second, a GA-based approach is proposed to generate interesting association rules without specified threshold of minimum confidence, taking advantage of the searching capability of GA. Finally, the redundant rules in the mining results are pruned by post-processing method. A test case is selected to demonstrate the feasibility and applicability of the proposed approach within the problem domain. It is concluded that the proposed method provided an effective tool to discover useful knowledge hidden in historical defect cases. The discovered knowledge indicating relationships between defects and defect causes enables project managers to make strategies for estimating and reducing defects.