Behavior data analysis for physical exercise through sports bracelets

Abstract

This paper proposes a systematic approach to analyze the learners’ behavior equipped with sports bracelets. The study begins with the step counts retrieved from the mobile database, transformed to behavior events and behavior states, which then are represented by consecutive strings of discretized step counts. Such behavior state strings are then decomposed and clustered to get behavior parameters: behavior lengths and behavior strengths. An integrated flow chart with behavior decomposition and behavior clustering algorithms are illustrated. The approach is implemented and verified by a five-month pilot experiment, through which 153379 step counts of 19 learners’ mobile sports bracelets are obtained. Henceforth, 2833 person-days behavior states with 72 time-periods are derived. Finally, these 19 learners are clustered into four groups: 9 for no exercise, 7 for average exercise, 2 for long-term exercise and 1 for very long-term exercise. Such behavior models may give more precise exercise planning in physical education.