Human Activity Recognition with Streaming Smartphone Data

• Published in: 2019 Global Conference for Advancement in Technology (GCAT) pp. 1 - 6
• Main Authors: Zhang, Yanjia, Ramachandran, Kandethody M.
• Format: Conference Proceeding
• Language: English
• Published: IEEE 01.10.2019
• Subjects: Decision Tree; Human Activity Recognition; Smartphone; Streaming Data; VFDT
• DOI: 10.1109/GCAT47503.2019.8978328

With the widely used smartphones, dynamic data coming from built in sensors, such as human activity data, can be easily obtained. Many applications' developments, such as applications in healthcare, fitness monitoring, and elder monitoring, are based on this kind of dynamic data. Although there are many offline methods that have made a great progress in analyzing these kinds of data, it still has a big challenge to get good results from a streaming data perspective. In this paper, we use an online method called Very Fast Decision Tree (VFDT) to mimic the real scenario. There are two main improvements from the existing models: 1) we train the model online and only use the examples data once for training instead of using them more than once; 2) after building VFDT, the model can be adjusted to identify new activities by adding only small amount of labeled observations. Our experiment on the same existing activities shows that the proposed algorithm achieves an average accuracy of 85.9% for all subjects and single subject accuracy rates are between 60.5% and 99.3%. Moreover, the average accuracy of learning new activity from a different data is 84% and single subject accuracy rate goes to as high as 100%.