Identification of human activities using mobile phone sensors based on stacking learning algorithm

Human Activity Recognition, as one of the growing fields of research, aims to identify which activity is done by individuals by tracking their activities. It has plenty of real-world applications such as health monitoring, abnormal behavior detection, and fitness supporting. Therefore, this study focuses on mobile phone data to distinguish and classify human activities by applying statistical features and using stacking learning method with the aim of improving the accuracy and precision of the classifiers. Two different stacking models are designed and applies on a public dataset. To show that stacking model performs much better than single classifiers in distinguishing each activity, classification results are compared with the result of two common single classifiers including Naïve Bayes and Random Forest. The results show that stacking methods can considerably improve classification accuracy, especially in the case of energetic activities that are difficult to distinguish, such as climbing stairs, walking, and jogging.

Keywords:

First keyword , Human Activity Recognition , Classification , Feature Extraction , Stacking Learning Algorithm