Semi-Automated Data Labeling for Activity Recognition in Pervasive Healthcare.

Area of Science:

• Pervasive Healthcare Monitoring
• Human-Computer Interaction
• Machine Learning for Sensor Data

Background:

• Accurate activity recognition models require labeled data, which is costly and time-consuming to obtain.
• Traditional data labeling methods are often performed in lab settings or with significant delays, impacting accuracy.
• Mobile and wearable sensors enable large-scale data collection in naturalistic environments.

Purpose of the Study:

• To develop and evaluate novel semi-automated online data labeling approaches for activity recognition.
• To address user burden and improve labeling accuracy in self-annotation.
• To facilitate more efficient and reliable data collection for healthcare monitoring.

Main Methods:

• Developed two semi-automated online labeling techniques for individuals performing activities.
• Method 1: Utilizes subtle finger gestures in response to labeling queries.
• Method 2: Employs an auditory activity classifier and conversational agent for clarification.

Main Results:

• Both novel approaches were evaluated in controlled experiments.
• The methods demonstrated feasibility and addressed key limitations of self-annotation.
• Achieved precision rates between 80% and 90%.

Conclusions:

• Semi-automated online labeling is a viable strategy for activity recognition data.
• The proposed methods offer improvements in user experience and data accuracy.
• Further research is needed to address the limitations identified in the studies.