Optimal Feature Set for Smartphone-based Activity Recognition

Human activity recognition using wearable and mobile devices is used for decades to monitor humans’ daily behaviours. In recent years as smartphones being widely integrated into our lives, the use of smartphone’s built-in sensors in human has been receiving more attention, which smartphone accelerometer plays main role. However, comparison standard machine, when developing a smartphone, limitations such processing capability energy consumption should be taken consideration, therefore, trade-off between performance computational complexity considered. this paper, we shed light on importance feature selection its impact simplifying classification process, enhances system. The novelty work related identifying most efficient features detection each individual uniquely. an experimental study with users different smartphones, investigated how achieve optimal set, system can decreased while accuracy remains high. For that, considered scenario, instructed participants perform activities, including static, dynamic, going up down stairs, walking fast slow freely holding their hands. To evaluate obtained set implementing two major algorithms, decision tree Bayesian network, activities. We further evaluated by comparing three sets from state-of-the-art. results demonstrated that replacing large number conventional only negligible overall it significantly decrease system’s complexity, essential smartphone-based systems.