ECG-Based Concentration Recognition With Multi-Task Regression

Area of Science:

• Human-computer interaction
• Biomedical engineering
• Machine learning

Background:

• Human activity and mental state recognition using wearables is a growing research area.
• Predicting stress levels via electrocardiogram (ECG) sensors is well-established.
• Accurate concentration prediction is challenging due to individual variability and high data sampling costs.

Purpose of the Study:

• To predict the degree of individual concentration using heart-rate features.
• To address challenges in building accurate concentration prediction models caused by individual diversity and high sampling costs.

Main Methods:

• Proposing a multitask learning (MTL) technique to share information among similar individuals.
• Conducting experiments with 18 healthy subjects performing daily office tasks.

Main Results:

• Demonstrating significant improvement in the accuracy of concentration prediction in small sample situations.
• Showing stable performance of the MTL method across different subjects, outperforming conventional models.

Conclusions:

• The proposed MTL method effectively improves concentration prediction accuracy, particularly in data-scarce scenarios.
• The stability and improved accuracy have significant implications for real-world concentration recognition systems.
• This approach drastically mitigates the data collection burden for individual users.