Confusion matrix for multiclass classification.

A Brain-Computer Interface (BCI) enables direct communication between the brain and external devices, such as computers or prosthetic limbs. This allows the brain to send commands while receiving sensory feedback from the device. Despite their potential, the performance limitations of existing BCI systems have motivated researchers to improve their efficiency and reliability. To address this challenge, the present study introduces a novel BCI paradigm centered on a cognitive task involving the reading of scrolling text in four different directions: right, left, up and down. The primary objective was to explore the electroencephalography (EEG) and near-infrared spectroscopy (NIRS) signals within this framework and assess the potential of hybrid BCI systems based on this innovative paradigm. The experimental protocol involved eight participants performing tasks across four classes of scrolling text. To optimize system accuracy and speed, EEG and NIRS data were segmented into discrete temporal windows. Features were extracted using the Hilbert Transform, while classification was performed via the k-nearest neighbor algorithm. The proposed approach achieved a classification accuracy of 96.28% 1.30% for multi-class tasks, demonstrating the effectiveness of hybrid modalities. This study not only introduces a novel paradigm for hybrid BCI systems, but also validates its performance, providing a promising direction for advancing the field.