Detecting Abnormal Behaviors in Smart Home

Recent popularity of smart home devices, due to the widespread use of Internet of Things (IoT) technology, has enabled infinite possibilities to provide people with more convenient, safe and healthy life experiences. Among various ideas and applications in smart home environments, the problem of abnormal behaviors detection arises people's attentions and concerns. In our work, we propose to achieve anomaly detection as a one-class classification problem. In particular, we use two machine learning models, which are Autoencoder and one-class Support Vector Machines, to learn the normal behavior patterns and then detect any behavior that deviates from the learned normality. We further use Samsung SmartThings, an open platform for smart homes and the consumer IoT, to build a real-time detection and notification mechanism, and evaluate effectiveness of our model in both offline and online detection scenarios.