Smartphone Sensor Data for Physical Activity Monitoring and Analysis

Abstract: As awareness of health risks associated with a sedentary lifestyle is raised, the demands fortechnological tools to help track one’s physical activity levels grow as well. A technologicalconcept that could be used to aid in this issue is human activity recognition (HAR), whichmaps signal data to recognisable human activity. The aim of this study is to explore thepossibilities of creating a method for utilising data collected from smart phone sensors totrack the type of human activity conducted (such as walking or running) and the quantity ofit measured in steps. This study consists of two main parts - classification and quantification.The classification part (HAR) looks at classifying the activities running, walking and cycling. Itis achieved through the machine learning algorithm SVM (support vector machine). Thequantification part comes in the form of a step counter that comes in two versions - onethat calculates the steps in a pre-recorded data, and another that is implemented as anAndroid application to calculate steps live. In both of these programs, the acceleration datais processed and the steps are then validated. The machine learning part achieved a finalaccuracy of 95% while the step counters scored 96% and 94%, respectively.