Hybrid CNN-Transformer Model for Accurate Classification of Human Attention Levels Using Workplace EEG Data

Accurately detecting human attention levels is a key challenge in cognitive neuroscience, with broad application value in improving productivity. Although Electroencephalography (EEG) signals are often used to study cognitive states, most studies still rely on data collected in controlled laboratory environments. This paper collects EEG data from employees during their daily work using a commercial single-channel EEG headband, making attention detection closer to real-world applications and increasing its feasibility and promotion value. We propose a new classification method based on a multi-head attention transformer to identify six different attention levels. We first perform a Short-Time Fourier Transform (STFT) on the EEG signal. Subsequently, we constructed a transformer architecture to effectively model long-range dependencies and subtle pattern changes in EEG data using self-attention and stacked encoder layers. Experimental results show that our proposed model achieves 87.37% classification accuracy in the six-level attention classification task, outperforming traditional high-performance methods and demonstrating superior performance compared to existing similar approaches. This achievement not only verifies the potential of the transformer architecture in EEG attention level classification but also provides new possibilities for developing advanced tools in fields such as brain-computer interface (BCI) and cognitive monitoring.