Reconstruction of Building Occupation by Machine Learning

The lack of knowledge regarding building uses currently constitutes a scientific barrier when it comes to predicting the energy consumption and comfort of buildings. One promising method for identifying uses in the existing buildings is to set up a measurement protocol associated with machine learning methods. In the context of applying machine learning methods to estimate occupancy, the input data X are conventionally constituted by numerous observations of different predictors such as temperature, sound or CO₂ concentration. The main difficulty in using supervised learning methods is that training the models requires labelled training data, that is data for which the true classification is known. To overcome this limitation, unsupervised learning methods may be considered, like hidden Markov models (HMMs). Although HMMs require training data, only the test data were considered in this study to evaluate its performance and to be able to compare it with that obtained by the random forest.