Comparison of different machine learning algorithms in predicting air conditioner operating behaviors in open-plan offices

Abstract

Some studies have applied machine learning (ML) methods to the field of occupant behavior, but the characteristics of different ML algorithms and their applicability are still unclear in predicting air conditioning (AC) behavior in open-plan office. Five ML algorithms, namely extreme gradient enhancement (XGBoost), logistic regression (LR), random forest (RF), support vector classification (SVC) and k-nearest neighbor (KNN), were applied to predict AC operating behavior in two open-plan office in Nanjing, China. The results showed that time of day, outdoor dry-bulb temperature, outdoor relative humidity and indoor CO2 concentration were the top four most important parameters in predicting AC behavior in open-plan offices. XGBoost and RF models had better accuracy and stability than other three models. The findings could shed light on the application of ML methods on the AC operating behavior modeling and other occupant behavior researches.