Mining event logs for knowledge discovery based on adaptive efficient fuzzy Kohonen clustering network

Abstract

As a digital representation of building projects, Building Information Modeling (BIM) can accumulate large volumes of log data containing hidden knowledge for deep exploration. However, such ever-increasing logs are likely to suffer from high complexity, inaccuracy, and uncertainty, which will inevitably raise challenges in uncovering latent and meaningful patterns. In order to yield satisfactory clustering quality and efficiency for better design process management, a novel clustering-based BIM event log mining approach is put forward in this paper. For one thing, a hybrid clustering algorithm named adaptive efficient fuzzy Kohonen clustering network (AEFKCN) is developed with a modified learning rate to accelerate the convergence. For another, a new clustering validity index (CVI) only relying on boundary points is designed to reduce computational complexity. An experiment is conducted in a 4 GB realistic BIM design event log dataset to validate the effectiveness of the proposed method. It begins from extracting a set of features associated with designers’ engagement and efficiency and ends up retrieving inherent insights into the person's design behavioral patterns. Moreover, the cluster analysis can significantly distinguish the design productivity at different time periods into the high, medium, and low level, which presents a unique opportunity in understanding and assessing design productivity objectively. Practically, our method can support data-driven decision making for managers to strategically schedule personalized work for different designers, aiming to boost design efficiency and smooth the design process.