A mobile system for sedentary behaviors classification based on accelerometer and location data

Abstract

BackgroundSedentary behaviors are associated to the development of noncommunicable diseases (NCD) such as cardiovascular diseases (CVD), type 2 diabetes, and cancer. Accelerometers and inclinometers have been used to estimate sedentary behaviors, however a major limitation is that these devices do not provide enough contextual information in order to recognize the specific sedentary behavior performed, e.g., sitting or lying watching TV, using the PC, sitting at work, driving, etc. ObjectivePropose and evaluate the precision of a mobile system for objectively measuring six sedentary behaviors using accelerometer and location data. ResultsThe system is implemented as an Android Mobile App, which identifies individual’s sedentary behaviors based on accelerometer data taken from the smartphone or a smartwatch, and symbolic location data obtained from Bluetooth Low Energy (BLE) beacons. The system infers sedentary behaviors by means of a supervised Machine Learning Classifier. The precision of the classification of five of the six studied sedentary behaviors exceeded 95% using accelerometer data from a smartwatch attached to the wrist and 98% using accelerometer data from a smartphone put into the pocket. Statistically significant improvement in the average precision of the classification due to the use of BLE beacons was found by comparing the precision of the classification using accelerometer data only, and BLE beacons localization technology. ConclusionsThe proposed system provides contextual information of specific sedentary behaviors by inferring with very high precision the physical location where the sedentary event occurs. Moreover, it was found that, when accelerometers are put in the user’s pocket, instead of the wrist and, when symbolic location is inferred using BLE beacons; the precision in the classification is improved. In practice, the proposed system has the potential to contribute to the understanding of the context and determinants of sedentary behaviors, necessary for the implementation and monitoring of personalized noncommunicable diseases prevention programs, for instance, sending sedentary behavior alerts, or providing personalized recommendations on physical activity. The system could be used at work to promote active breaks and healthy habits.