Physical Activity Monitoring and Classification Using Machine Learning Techniques.

Area of Science:

• Human-Computer Interaction
• Biomedical Engineering
• Machine Learning

Background:

• Physical activity is vital for obesity control and healthy living, especially with pandemic-related restrictions.
• Wearable sensors and machine learning (ML) offer promising tools for tracking physical activities and promoting healthy lifestyles.
• Daily life activities exhibit inherent class imbalance, with some activities occurring more frequently than others.

Purpose of the Study:

• To investigate the impact of class imbalance on ML classifier performance for physical activity recognition.
• To identify which ML classifiers are most sensitive to class imbalance in sensor data.
• To develop novel techniques for accurate physical activity logging using wearable sensors and ML.

Main Methods:

• Utilized motion sensor data from 30 participants performing various daily activities.
• Introduced class imbalance by using different training data splits.
• Evaluated the performance of state-of-the-art ML algorithms under varying degrees of data imbalance.

Main Results:

• Class imbalance significantly affects the performance of ML systems for physical activity recognition.
• Underrepresentation of certain physical activities during training substantially impacts classifier accuracy.
• Specific ML classifiers demonstrated varying sensitivity to class imbalance.

Conclusions:

• Addressing class imbalance is critical for developing robust and accurate physical activity recognition systems.
• Future research should focus on strategies to mitigate the effects of class imbalance in wearable sensor data.
• Effective physical activity tracking systems are essential for public health initiatives targeting obesity and sedentary lifestyles.