Background: Among the factors that outline the health of populations, person's lifestyle is the more important one. This work focuses on the caracterization and prevention of sedentary lifestyles. A sedentary behavior is defined as "any waking behavior characterized by an energy expenditure of 1.5 METs (Metabolic Equivalent) or less while in a sitting or reclining posture".
Objective: To propose a method for sedentary behaviors classification using a smartphone and Bluetooth beacons considering different types of classification models: personal, hybrid or impersonal.
Results: Following the CRISP-DM methodology, a method based on a two-layer approach for the classification of sedentary behaviors is proposed. Using data collected from a smartphones' accelerometer, gyroscope and barometer; the first layer classifies between performing a sedentary behavior and not. The second layer of the method classifies the specific sedentary activity performed using only the smartphone's accelerometer and barometer data, but adding indoor location data, using Bluetooth Low Energy (BLE) beacons. To improve the precision of the classification, both layers implemented the Random Forest algorithm and the personal model.
Conclusions: This study presents the first available method for the automatic classification of specific sedentary behaviors. The layered classification approach has the potential to improve processing, memory and energy consumption of mobile devices and wearables used.
Keywords: Machine learning; activities of daily living; sedentary behavior.