Machine Learning Algorithm for Inferring User Activities from IoT Device Events

The ubiquitous and heterogeneous deployment of internet of things (IoT) devices in smart homes has created new opportunities to extract knowledge, awareness, and intelligence via monitoring and understanding the devices’ interactions with their environments and users.  Meaningful information from network traffic of IoT devices in smart homes can be discoverable even though much of the traffic is often encrypted over secure wireless networks or via IoT application-level encryption.  Current solutions for inferring user activities from IoT devices (or reconstructing real-world user activities in smart homes and matching them with ground-truth) requires an exact sequence of user activities, rather than a multiset of user activities.  There is a need to concentrate on a small number of representative matches of user activity patterns and design an efficient algorithm for computing these representatives.

Researchers at Arizona State University (ASU) have developed an algorithm for inferring user activities from IoT device events.  This algorithm solves the problem of inferring a sequence of user activities together with their patterns from a sequence of device events in a smart home setting.  One significant contribution of the algorithm is the unsupervised learning aspect which helps make the inference more adaptive to varying scenarios.  Essentially, user activity patterns are inferred from a sequence of device events by first deterministically extracting a small number of representative user activity patterns from the sequence of device events, then applying unsupervised learning to compute an optimal subset of these user activity patterns to infer user activity.

Conducted experiments with both real and synthetic data and demonstrated that ASU algorithm is robust and outperforms the state-of-the-art solution.  For example, based on extensive experiments with sequences of device events triggered by 2,959 real user activities and up to 30,000 synthetic user activities, the ASU algorithm is resilient to device malfunctions and transient failures/delays.

Related Publications: An Effective Machine Learning Based Algorithm for Inferring User Activities From IoT Device Events

 Potential Applications:

  • Smart Home Technology
  • Security and Surveillance (e.g., home safety monitoring)
  • Healthcare and Elderly Care (e.g., assisted living care)

Benefits and Advantages:

  • Enhancing Home Safety
  • Empowering Assisted Living
  • Optimizing IoT Device Synergy