Abstract

Classification and modeling of activity duration provide significant characteristics to estimate the psychological behaviour of a smart home resident. This article validates the fact, which was only an assumption previously, that smart home event duration can be modeled in Gaussian distribution. It proposes a temporal prediction algorithm based on Gaussian distribution to predict the duration of an event interval, which approximates the ending time of the smart home user’s activities. It incrementally estimates the ending time of an event that follows the central limit theory of statistical probability. The results and analysis imply that temporal duration follows Gaussian distribution, which expresses almost the same property of Gaussian equation. The algorithm is verified with significant amount of MavHome and MIT PlaceLab smart home sensory data, which exhibit 88.3 and 90.3% prediction accuracies respectively. Finally, the proposed temporal algorithm is utilized for temporal anomaly detection, which has detected 54 and 46 abnormal behaviour when tested with MavLab and MIT PlaceLab data respectively.

Key words: Temporal duration, smart homes, Gaussian distribution, prediction algorithm, temporal prediction, anomaly detection.