Hybrid Convolutional-Gated Recurrent Neural Network for Robust Mobile Health Activities Classification

Area of Science:

• Biomedical Engineering
• Artificial Intelligence in Healthcare
• Mobile Health Technology

Background:

• Mobile health (mHealth) enables patient monitoring of body activities and vital signs via devices like smartphones and smartwatches.
• Artificial Intelligence (AI) is increasingly used in healthcare for automated diagnosis, especially in real-time emergency situations.

Purpose of the Study:

• To introduce a novel deep learning model for automatic identification of daily human physical activities.
• To develop and validate a hybrid AI model for robust human activity classification using mobile sensor data.

Main Methods:

• A hybrid Convolutional Neural Network (CNN) and Gated Recurrent Unit (GRU) deep learning model was designed.
• The CNN-GRU model was trained and evaluated on the public MHEALTH dataset, comprising 12 distinct physical activities.
• Performance was benchmarked against existing machine learning and deep learning classifiers, including CNN-LSTM and 1-D CNN.

Main Results:

• The proposed CNN-GRU model achieved a state-of-the-art classification accuracy of 99.50% on the MHEALTH dataset.
• The model demonstrated superior performance compared to CNN-LSTM (98.83%), 1-D CNN (96.89%), and ensemble methods.
• High precision, recall, and F1-scores were maintained across all 12 activity classes.

Conclusions:

• The developed CNN-GRU model offers a highly accurate and robust solution for human activity recognition using mobile health data.
• This AI model has significant potential for implementation in cloud-based systems for continuous health monitoring, particularly for senior patients.
• The findings highlight the effectiveness of combining CNN for spatiotemporal feature extraction and GRU for temporal dependency modeling in mHealth applications.