Activity Data Cluster Analysis

Discovering a pattern in someone's daily physical activity.

Identifying and displaying a daily pattern of physical activity is a hard problem because two days may have the same overall pattern, but one day may be a non-uniformly compressed and shifted version of the other day. The base data for this project was several years worth of raw data on individuals’ minute-by-minute activity levels each day. After exploring analysis techniques in R, a process involving dynamic time warping, LOESS smoothing, and unsupervised hierarchical clustering techniques to recognize and display an individual’s most common daily patterns proved the best solution. The project culminated in designing and implementing a web app for users to explore their data on a mobile device.

Raw activity data overlayed from multiple days. Actual results can't be displayed for confidentiality reasons.