Supervised learning-based assessment of office layout satisfaction in academic buildings

Employee satisfaction significantly affects health, well-being and productivity, and office layout plays a dominant role in office psychological satisfaction. However, existing studies have not yet proposed a quantitative evaluation method for office layout satisfaction to assist design decisions. This study conducts a post-occupancy evaluation (POE) process of office layout satisfaction from 1,317 staff members at 3 universities in the Yangtze River Delta, China. The proposed office layout feature network supports the questionnaire design and environmental measurement. Based on the survey data, multiple resampling methods are considered to face the imbalanced dataset problem, and feature selection integrates statistical analysis methods and machine learning algorithms. Nine supervised learning algorithms are tested for office layout satisfaction prediction, and the final predictive model is established based on the random forest algorithm. The predictive model explanation is further integrated with original data analysis to extract the quantified impacts of various building characteristics. The workstation adjustment under the background of COVID-19 in an actual staff office is chosen to be an application scenario of the predictive model. The results show that the workstation distance, room depth and room width-depth ratio are dominant in the evaluation of office layout satisfaction. The proposed predictive model achieves 64.5% accuracy, and the prediction results are interpretable, which promotes its application in office design practice. The data processing methods in this study respond to the common data problems in the POE based opinion collection process. The extracted influence mechanisms of building characteristics can directly support user-centered office design.