Using a data driven neural network approach to forecast building occupant complaints

Abstract

Occupant complaints are a reflection of poor building performance and an unsatisfactory indoor environment. One way to mitigate these complaints and to ensure occupants' satisfaction with building performance is through a well-performing facility management that is capable of planning for and addressing the maintenance services. This research work proposes a machine learning-based multistep generic framework to analyze building occupant complaints and to forecast the number of thermal complaints, in particular, as part of the facility management's predictive maintenance strategy. The developed forecasting model is benchmarked against a traditional statistical model to ensure a satisfactory performance. The proposed methodology was tested for a period of three years on a highly unstructured and unsolicited occupant-complaint data recorded by facility management operators in a residential complex composed of 16 buildings. Text mining results of more than 6,000 complaints showed that thermal complaints are among the most common ones thus require further attention of facility managers. The developed Multi-Layer Perceptron (MLP) models to forecast the number of thermal complaints for the upcoming week showed similar or better performance as compared to the traditional Autoregressive Integrated Moving Average (ARIMA) models with a higher ability to generalize to new data. For example, the cooling related complaints MLP model showed a RMSE of 0.878, which is 21% lower than the RMSE of the ARIMA model. It is also evident that the developed models could assist facility managers in planning for the staffing resources required to handle these complaints thus enhancing occupants' satisfaction and the building performance. © 2021 Elsevier Ltd