Dynamic clustering and modeling of temporal data subject to common regressive effects

• Published in: Neurocomputing (Amsterdam) Vol. 500; pp. 217 - 230
• Main Authors: Bonfils, Louise, Samé, Allou, Oukhellou, Latifa
• Format: Journal Article
• Language: English
• Published: Elsevier B.V 21.08.2022; Elsevier
• Subjects: Clustering; Dynamic latent variable model; Engineering Sciences; Occupant behavior in buildings; Other; Urban data; Variational inference approximation; Urban data; Occupant behavior in buildings; Clustering; Variational inference approximation; Dynamic latent variable model; DYNAMIC LATENT VARIABLE MODEL; VARIATIONAL INFERENCE APPROXIMATION; CLUSTERING; URBAN DATA; OCCUPANT BEHAVIOR IN BUILDINGS
• ISSN: 0925-2312; 1872-8286
• DOI: 10.1016/j.neucom.2022.05.038

•Clustering model with dynamic latent variables and common regressive effects.•The variational inference method is used to estimate the model parameters.•Performance evaluation performed on simulated data and compared to reference models.•Application on real indoor temperature data highlights different occupancy behaviors. Clustering is used in many applicative fields to organize data into a few groups. Motivated by behavioral extraction issues from urban data, this paper proposes a new clustering method to model clusters with dynamic profiles while considering common regressive effects. As maximum likelihood estimation is not suitable in this case, the parameters of the proposed model were estimated using variational approximation. The ability of the model to estimate parameters was evaluated using various simulated data and compared with two other models. The article also proposes an application of this model to the extraction of occupant behavior in buildings using a real open source indoor temperature database. The objective is to classify individual houses according to indoor temperature while estimating the effect of meteorological variables and class profiles that can be interpreted as occupancy behaviors.