Transforming wearable sensor data for robust feature selection in human activity recognition using reinforcement...

Area of Science:

• Biomedical Engineering
• Artificial Intelligence in Healthcare
• Wearable Sensor Technology

Background:

• Smart healthcare systems increasingly rely on body sensor data.
• Existing models struggle with large datasets and accurate classification.
• Challenges include capturing complex temporal patterns in sensor readings.

Purpose of the Study:

• To develop an advanced deep reinforcement learning model for enhanced body sensor data analysis.
• To improve the accuracy and robustness of activity recognition in smart healthcare.
• To address limitations in current models for handling massive, noisy datasets.

Main Methods:

• Utilized time-sequential data and the Generative Actor-Critic (GAC) deep reinforcement learning technique.
• Integrated cyclic Generative Adversarial Networks with GAC for robust activity modeling.
• Employed wearable sensor data collection to enhance feature selection by improving inter-class differences and decreasing intra-class variations.

Main Results:

• Achieved high accuracy in activity recognition, reaching 98.76% on the UCI-HAR dataset and 98.84% on the Motion Sense dataset.
• Demonstrated superior performance compared to traditional deep learning techniques, particularly in noisy conditions.
• Successfully enhanced feature selection for wearable sensor data, leading to more robust models.

Conclusions:

• The proposed deep reinforcement learning approach, integrating GAC and cyclic GANs, offers a significant advancement in smart healthcare data analysis.
• This method provides accurate and robust activity recognition from wearable sensor data, outperforming existing techniques.
• The findings highlight the potential of advanced AI for improving the efficacy of smart healthcare systems.