Cohort studies are a type of longitudinal research. They follow the same group of people throughout their lives, charting social change and untangling the reasons behind it. When the surveys are complete, the data is cleaned and documented and deposited for use by the scientific community.
In this project we look at data from the Millenium Cohort as organised by the Centre of Longitudinal Studies in London (Institute of Education, UCL). Over 6000 adolescent members of the Millenium Cohort wore a wearable movement sensor (accelerometer) on their wrist for two days. During these two days participants filled in an activity log at a 10-minute resolution.
Our aim is to extract insight from the sensor data beyond the results from existing heuristic algorithms, while trying to avoid over-interpretation given the uncertainties that are introduced by uncontrolled real life experimental conditions. We develop an unsupervised data-driven model for identifying clusters in the accelerometer time series data. To aid interpretation we fuse the model output with the activity diary recordings as well as conventional heuristic algorithm output.
Image: Exercise by Katchooo