Abstract
Introduction: Attention is a complex cognitive function of the human brain that plays a vital role in our daily lives. Electroencephalogram (EEG) is used to measure and analyze attention due to its high temporal resolution. Although several attention recognition brain-computer interfaces (BCIs) have been proposed, there is a scarcity of studies with a sufficient number of subjects, valid paradigms, and reliable recognition analysis across subjects.
Methods: In this study, we proposed a novel attention paradigm and feature fusion method to extract features, which fused time domain features, frequency domain features and nonlinear dynamics features. We then constructed an attention recognition framework for 85 subjects.
Results and discussion: We achieved an intra-subject average classification accuracy of 85.05% ± 6.87% and an inter-subject average classification accuracy of 81.60% ± 9.93%, respectively. We further explored the neural patterns in attention recognition, where attention states showed less activation than non-attention states in the prefrontal and occipital areas in $\alpha, \beta$ and $\theta$ bands. The research explores, for the first time, the fusion of time domain features, frequency domain features and nonlinear dynamics features for attention recognition, providing a new understanding of attention recognition.
Introduction
While Brain-Computer Interfaces (BCIs) have made significant progress in motor control and communication, reliable attention recognition remains a major challenge in neuroscience. Most existing studies suffer from three primary limitations: small sample sizes (often fewer than 10 subjects), inconsistent experimental paradigms to induce focus, and poor "inter-subject" performance—meaning a model trained on one individual rarely generalizes to another. Accurate, real-time attention monitoring is critical for high-stakes environments like military operations, preventing fatigue-related driving accidents, and clinical applications for ADHD or depression. This study seeks to bridge the gap between lab-based EEG analysis and a generalized framework that works reliably across a diverse population.
Methods
Experimental Paradigm
The researchers introduced a novel experimental paradigm to induce clear "attention" and "non-attention" states. The study utilized a dataset of 85 healthy subjects (45 male, 40 female, average age 25.3), representing one of the larger cohorts in the field. Data was recorded using a 32-channel Neuroscan system at a 250Hz sampling rate.
To induce a focused state, subjects performed a mental arithmetic task, such as continuously subtracting 3 from a random number appearing on a screen. The "non-attention" state was defined as a simple open-eyed resting state.
Feature Fusion Method
Instead of relying on a single data type, the team extracted 123 features per EEG channel, categorized into three domains:
- Time Domain: 12 features including Hjorth parameters (activity, mobility, complexity) and statistical measures like skewness and kurtosis.
- Frequency Domain: 48 features, primarily Power Spectral Density (PSD) across standard brain rhythms ($\delta, \theta, \alpha, \beta, \gamma$) and wavelet coefficients.
- Non-linear Dynamics: 63 features designed to capture the complexity and chaos of brain signals, such as various types of Entropy (Sample, Renyi, Shannon) and Fractal Dimensions.
These features were processed using three standard machine learning classifiers: Random Forest (RF), Decision Tree (DT), and Support Vector Machine (SVM).
Results and Discussion
Classification Accuracy
The proposed fusion method demonstrated superior performance over baseline methods. The core finding was that fusion features significantly outperformed individual feature sets, with non-linear dynamics providing the "missing link" for inter-subject reliability.
- Intra-subject Accuracy: 85.05% ± 6.87% (using Random Forest).
- Inter-subject Accuracy: 81.60% ± 9.93% (using SVM).
The fusion approach outperformed time-domain, frequency-domain, and non-linear features used in isolation across all classifiers ($p < 0.01$).
Neural Patterns of Focus
Topographical maps revealed a surprising pattern: during the attention state, there was actually less activation (lower power) in the prefrontal and occipital areas within the $\alpha, \beta,$ and $\theta$ bands compared to the resting state. This suggests that "focus" may involve the brain suppressing certain background rhythms to process task-specific information more efficiently. While many features work well for individuals, only a subset—specifically PSD in the $F_4$ band, Wavelet Entropy, and Teager-Kaiser energy—remained highly effective for inter-subject classification.
Conclusion
This research provides a robust framework for EEG-based attention recognition. Despite the high accuracy, future work is needed to transition from offline analysis to real-time, closed-loop BCI systems. Key areas for development include channel reduction to identify the minimum number of electrodes needed for wearable technology and the exploration of end-to-end deep learning architectures to further improve recognition without manual feature selection.
Summary updated