Toward Concurrent Identification of Human Activities with a Single Unifying Neural Network Classification: First Step

Area of Science:

• Health Informatics
• Artificial Intelligence in Healthcare
• Behavioral Science

Background:

• Understanding human behavior is key to modeling health.
• Wearable technology allows for unobtrusive monitoring of daily activities.
• Quantifying behavior passively is crucial for linking it to health outcomes.

Purpose of the Study:

• To develop and validate an AI-driven method for recognizing human health behaviors using smartwatch data.
• To assess the performance of a deep neural network in segmenting time-series activity data.
• To explore the potential of AI in healthcare for diagnosis, prognosis, and intervention.

Main Methods:

• Sixty adult participants emulated specific behaviors (smoking, exercise, eating, medication intake) in a lab.
• Smartwatches captured accelerometer data during these activities.
• A deep neural network (CNN-LSTM) was trained on annotated data for activity segmentation.

Main Results:

• The AI model achieved an average macro-F1 score of at least 85.1% via leave-one-subject-out cross-validation.
• This demonstrates high performance in distinguishing between various human behaviors.
• The findings highlight the method's potential for real-world health applications.

Conclusions:

• AI, particularly deep learning with sensor data, can effectively characterize human health behaviors.
• This technology offers a promising tool for early healthcare interventions and personalized patient care.
• AI can support healthcare professionals in making data-driven decisions for improved health outcomes.