NICOP - Discovering Periodic Human Behavior from Multivariate Sensor Data
Abstract
Nowadays almost everyone is carrying a smartphone, and the smartphones are producinghuge amounts of personal big data. Typically, a"" smartphone is equipped with varioussensors, including GPS, accelerometer, and gyroscope. Thanks to these sensors, it hasbecome ea""sy to collect various user activity data, and these data are really useful materialsfor modeling the behavior of the user who carri"es the smartphone. The goal of this projectis to understand routine human behavior more naturally and precisely from multivariates"ensor data, typically obtained via smartphones. First, we define a novel concept of themultivariate conditional periodic pattern (M""CPP) for multivariate streams. Next, wedevelop an efficient and light-weighted algorithm of finding these patterns. Then, based on""the patterns discovered by the algorithm, we attempt to improve the accuracy of predictinghuman behavior such as interruptibility."" Last, in order to study long-term human behavior,we try to find the points when the patterns change or drift and associate such po""ints withreal-world events. Overall, we expect that our approach will improve the detection ofroutine behavior compared with the s""tate-of-the-art approaches. The outcome of thisproject can be used in various applications which exploit such routineness, e.g., in""telligentassistance, healthcare, and public transportation services.
Document Details
- Document Type
- DoD Grant Award
- Publication Date
- Jan 23, 2018
- Source ID
- N629091812054
Entities
People
- Jae-Gil Lee
Organizations
- KAIST
- Office of Naval Research
- United States Navy