Deep Learning in the Recognition of Activities of Daily Living Using Smartwatch Data

Area of Science:

• Computer Science
• Artificial Intelligence
• Wearable Technology

Background:

• Human Activity Recognition (HAR) using wearable devices like smartwatches is a growing field in computer science.
• Understanding daily activities provides valuable insights into individual behavior and health.

Purpose of the Study:

• To comparatively evaluate deep learning techniques for recognizing Activities of Daily Living (ADL).
• To identify the most effective deep learning architecture for HAR from smartwatch data.

Main Methods:

• A literature review mapped existing HAR techniques.
• Three deep learning techniques were selected and evaluated on a dataset.
• Performance was assessed using standard metrics: accuracy, precision, recall, and F1-score.

Main Results:

• The DeepConvLSTM architecture demonstrated superior performance in ADL recognition.
• This architecture integrates recurrent convolutional layers with a single LSTM layer.

Conclusions:

• DeepConvLSTM shows significant promise for accurate HAR in real-world applications.
• Software leveraging this architecture can enhance smartwatch users' understanding of their movement patterns.