Development of building defects dashboards and stochastic models for multi-storey buildings in Victoria, Australia

Abstract

Purpose Building defects are becoming recurrent phenomena in most high-rise buildings. However, little research exists on the analysis of defects in high-rise buildings based on data from real-life projects. This study aims to develop dashboards and models for revealing the most common locations of defects, understanding associations among defects and predicting the rectification periods. Design/methodology/approach In total, 15,484 defect reports comprising qualitative and quantitative data were obtained from a company that provides consulting services for the construction industry in Victoria, Australia. Data mining methods were applied using a wide range of Python libraries including NumPy, Pandas, Natural Language Toolkit, SpaCy and Regular Expression, alongside association rule mining (ARM) and simulations. Findings Findings reveal that defects in multi-storey buildings often occur on lower levels, rather than on higher levels. Joinery defects were found to be the most recurrent problem on ground floors. The ARM outcomes show that the occurrence of one type of defect can be taken as an indication for the existence of other types of defects. For instance, in laundry, the chance of occurrence of plumbing and joinery defects, where paint defects are observed, is 88%. The stochastic model built for door defects showed that there is a 60% chance that defects on doors can be rectified within 60 days. Originality/value The dashboards provide original insight and novel ideas regarding the frequency of defects in various positions in multi-storey buildings. The stochastic models can provide a reliable point of reference for property managers, occupants and sub-contractors for taking measures to avoid reoccurring defects; so too, findings provide estimations of possible rectification periods for various types of defects.