| dc.description.abstract | Advancements in technology have led to more intelligent computing systems, butserver-based computing presents itself with network delay and high energy consumptionleading to significant carbon footprints. A lag in computing the resultand rendering it on a user�s end device, such as a smartphone could lead to poorquality of experience or user experience. Edge computing and its variants suchas fog and cloudlet computing, are paradigms for processing data close to thesource. However, the computing platform still remains to be a high-end server.The pipeline of mobile (smartphone-based) sensing involves sensing, data preprocessing,feature extraction, model training, and testing. The majority of workimplemented only the sensing phase whereas the remaining were offloaded to aserver. However, smartphones have rich computing power in terms of octa-coreprocessors, faster clock speeds that are under-utilized. Few attempts have beenmade to utilize this computing power. Benchmark studies have shown that theperformance of smartphone processors equals an Intel i3 processor. Most importantly,these devices consumed 30 times lesser energy than traditional servers forthe same computing task.In this work, we design, develop and implement a mobile application wherethe entire pipeline is implemented on the phone. The application takes the collectedsensor data on which we implement machine learning algorithms to classifythe physical activity of a user. Performance results show that the app consumes350mAh amount of battery, with a CPU utilization of 13%. We also receivedan average user rating of 4.5/5 for user experience on the impact of positive interventionsthat our app automatically provides. This thesis provides a frameworkfor implementing applications on smartphones eliminating the need for offloading. | |
