Deep Learning for Classifying Physical Activities from Accelerometer Data

Area of Science:

• Biomedical Engineering
• Artificial Intelligence
• Health Informatics

Background:

• Physical inactivity is a major risk factor for non-communicable diseases like heart disease, diabetes, and cancer.
• Current methods for monitoring patient physical activity are limited in precision and scope.
• Accurate tracking of physical activity is crucial for effective disease prevention and management.

Purpose of the Study:

• To develop and evaluate an artificial intelligence (AI)-based system for classifying physical movement activity patterns.
• To improve the accuracy and detail of physical activity monitoring for medical and training purposes.
• To provide healthcare professionals with a tool for precise patient physical activity assessment.

Main Methods:

• Utilized two deep learning (DL) models: a deep feed-forward neural network (DNN) and a deep recurrent neural network (RNN).
• Evaluated models on two distinct physical movement datasets comprising activities of daily life (ADL).
• Datasets were collected from volunteers wearing tri-axial accelerometer sensors on wrists and hips, encompassing 14 and 10 ADLs respectively.

Main Results:

• The deep recurrent neural network (RNN) model demonstrated superior performance in classifying fundamental movement patterns.
• The RNN model achieved an overall accuracy of 84.89% and an F1-score of 82.56%.
• Results indicate the AI approach significantly outperforms existing state-of-the-art methods in activity classification.

Conclusions:

• The proposed AI-driven physical activity classification system offers a promising solution for precise patient monitoring.
• This technology can empower medical doctors and personal trainers with detailed insights into patient physical activities.
• Enhanced understanding of patient movement patterns can lead to more personalized and effective treatment strategies.