Modeling Users’ Data Traces in Multi-Resident Ambient Assisted Living Environments

Abstract

Modeling users’ data traces is of crucial importance for human behavior analysis and context-aware applications in ambient assisted living (AAL) environments. However, learning the parameters of the underlying model is a challenging task in multi-occupant environments; because, the anonymous users’ data traces are aggregated temporally. This paper proposes a novel method for modeling users’ data traces in multi-resident sensor-based AAL environments. A Markov chain was considered as the underlying model. We aimed at estimating the parameters of the Markov chain directly out of users’ aggregate data. For this purpose, we hired the idea of conditional least squares (CLS) estimation. However, the CLS estimations can be inconsistent in the circumstances of AAL environments. To tackle this problem, we proposed to regularize the CLS estimations using spatial information of sensors. This information was extracted using an accessibility graph, made out of the deployed sensor network. To evaluate the proposed method, a well-known and publicly available dataset was used. The proposed method was compared with the standard CLS, using Kullback-Leibler (KL) divergence, and mean squared error (MSE) criteria. The results conveyed that the proposed method results in estimations with lower KL divergences from ground truth, compared to CLS. Also, the proposed method outperformed CLS with a MSE of 2.7 × 10−3.

Keywords

• Ambient assisted living (AAL)
• smart environments
• conditional least squares (CLS) estimation
• aggregate data
• Markov chain
• multi-resident environments