Unsupervised Anomaly Detection Using Light Switches for Smart Nursing Homes
Chih-Wei Ho, Chun‐Ting Chou, Yu-Chun Chien, Chia-Fu Lee · 2016
Anomaly detection plays a critical role in various smart living scenarios. However, achieving effective detection while not imposing burdens on users is never an easy task. To solve this problem, an unsupervised anomaly detection algorithm using light switches is proposed. By using an unsupervised approach, care takers in nursing homes do not need to label the collected data. By using information generated by smart switches, senior citizens are not forced to use wearables, change the battery or feel privacy invasion from cameras. Our solution adopts the statistical-based algorithm based on expectation maximization (EM). By adding constrains to reduce the high variances of the mixture model and recursively removing the extremely-low probability data from the model, a more accurate mixture model can be constructed. Our experiments in a real apartment show that the false alarm rate can be reduced by at least 56% compared to the existing cluster-based algorithms when the targeted miss detection rates are low.