A Case Study to Examine the Imputation of Missing Data to Improve Clustering Analysis of Building Electrical Demand

Building performance data are widely used for daily operation, improving building efficiency, identifying and diagnosing performance problems, and commissioning. In this study, the authors explore the use of missing data imputation and clustering on an electrical demand dataset. The objective was to compare four approaches of data imputation and clustering analysis. Results of this study suggest that using multiple imputation to fill in missing data prior to performing clustering analysis results in more informative clusters. Commonly used methods to fill in missing data lead to changes in cluster membership that are not suggestive of a change in the building's performance, but instead is a result of the choice of imputation method used. Practical application: The authors demonstrate, through the use of a case study, the application of a statistically sound method for filling in missing data in large buildings performance datasets. The methods used in this analysis are available through the open-source programming language R and are straight forward to implement. The approach demonstrated in this case study could aid buildings analysts with fault detection and continuous commissioning of large commercial buildings.