HARNet: automatic recognition of human activity from mobile health data using CNN and transfer learning of LSTM with SVM

Abstract

Human Activity Recognition (HAR) system is analysing human behaviour using mobile health technology. Mobile Health data (MHEALTH) uses electronic devices to collect data and identify the activity of the patient in real-time. Recordings of 10 patients’ vital signs from various circumstances are included in the dataset. With a sensor attached to their bodies, they were required to carry out a number of physical tasks. Due to the lack of accuracy in the other state-of-the-art algorithms, we proposed Human Activity Recognition Neural Network (HARNet) architecture for automatic recognition of human activity using CNN and LSTM with the transfer learning of SVM. Here, the human health behaviour was analysed and classified using different ML and DL algorithms. The hybrid techniques of CNN and LSTM are selected across the different DL algorithms and it is used to extract independent and discriminating features, which aids the SVM classifier to attain good classification. When compared to other DL methods, HARNet performed better, achieving 99.8% accuracy. Overall, HAR systems have many potential applications in various fields, including healthcare, wellness, sports and surveillance. They have relationships to many different academic disciplines, including sociology, human–computer interaction, medical and may offer individualized help for different domains.

Keywords

• Human Activity Recognition
• Mobile Health data (MHEALTH)
• Convolutional Neural Network (CNN)
• Long Short Term Memory (LSTM)
• Support Vector machine (SVM)
• deep learning