Modeling individual complaint behavior in daily office environment using a novel one-class, multi-linear classifier

Learning the model of user´s thermal complaints in the daily office environment and apply it to the control of Heating, Ventilating, and Air Conditioning (HVAC) is more user-friendly and intelligent. But the modeling of complaint behavior is challenging because of the stochastic complaint time, unbalanced sample sets, individual differences and uncertainties. Most existing human comfort models are not feasible under those conditions. In this paper, we propose a novel one-class (complaint only) classifier to model individual human complaints. The method extracts the pare-to-frontier set of the samples of each individual and uses a multi-linear classifier to describe the boundary of the complaint region in parameter space. Virtual no-complaint samples are synthesized to calculate the classifiers´ false positive rate. Real experimental results show that the method have lower false negative rate than traditional classifiers, which have more significant value in avoiding complaints. We also propose a method to do trade-off between false negative rate and false positive rate.