A feasibility study on smartphone accelerometer-based recognition of household activities and influence of smartphone position

BACKGROUND: Obesity and physical inactivity are the most important risk factors for chronic diseases. The present study aimed at (i) developing and testing a method for classifying household activities based on a smartphone accelerometer; (ii) evaluating the influence of smartphone position; and (iii) evaluating the acceptability of wearing a smartphone for activity recognition. METHODS: An Android application was developed to record accelerometer data and calculate descriptive features on 5-second time blocks, then classified with nine algorithms. Household activities were: sitting, working at the computer, walking, ironing, sweeping the floor, going down stairs with a shopping bag, walking while carrying a large box, and climbing stairs with a shopping bag. Ten volunteers carried out the activities for three times, each one with a smartphone in a different position (pocket, arm, and wrist). Users were then asked to answer a questionnaire. RESULTS: 1440 time blocks were collected. Three algorithms demonstrated an accuracy greater than 80% for all smartphone positions. While for some subjects the smartphone was uncomfortable, it seems that it did not really affect activity. CONCLUSIONS: Smartphones can be used to recognize household activities. A further development is to measure metabolic equivalent tasks starting from accelerometer data only.