Human Activity Recognition Using Time Series Pattern Recognition Model-Based on Tsfresh Features

Wearable devices are increasingly used to monitor people's activities, so data acquired from sensors are more available to establish models for recognizing human activities. This paper proposes a TSPR-model that can extract time-series features from sensor data by tsfresh in python. The model, which is less sensitive to data from different people, is able to mine the characteristics of sensor data and solve the noise of on-site data. Besides catering to the actual need in daily life, we perform subject-dependent and subject-independent evaluations. The accuracy has dramatically improved, especially in the latter, indicating that the proposed model is robust in distinctive subjects. Finally, due to different accelerometer patterns in different sexes and age groups shown in figures, we propose models based on clustered samples according to personal characteristics. They outperform models based on total samples.