EF-Net is a CNN-based multimodal deep learning model designed for mental state recognition using EEG and fNIRS data. The paper evaluates EF-Net on the EEG-fNIRS word generation (WG) dataset, focusing on subject-independent settings. The model outperforms five baseline approaches, including traditional machine learning and deep learning methods, achieving F1 scores of 99.36%, 98.31%, and 65.05% in subject-dependent, subject-semidependent, and subject-independent settings, respectively. EF-Net excels in extracting temporal features from EEG and spatial features from fNIRS, combining them for effective mental state classification. The model demonstrates strong performance across different data usage scenarios and training-testing splits, highlighting its capability to generalize across unseen subjects. The study underscores the effectiveness of integrating EEG and fNIRS data for brain activity analysis, with EF-Net showing significant improvements in accuracy and F1 scores compared to existing methods. The results suggest that EF-Net has potential for real-world applications in mental state recognition and brain activity learning.EF-Net is a CNN-based multimodal deep learning model designed for mental state recognition using EEG and fNIRS data. The paper evaluates EF-Net on the EEG-fNIRS word generation (WG) dataset, focusing on subject-independent settings. The model outperforms five baseline approaches, including traditional machine learning and deep learning methods, achieving F1 scores of 99.36%, 98.31%, and 65.05% in subject-dependent, subject-semidependent, and subject-independent settings, respectively. EF-Net excels in extracting temporal features from EEG and spatial features from fNIRS, combining them for effective mental state classification. The model demonstrates strong performance across different data usage scenarios and training-testing splits, highlighting its capability to generalize across unseen subjects. The study underscores the effectiveness of integrating EEG and fNIRS data for brain activity analysis, with EF-Net showing significant improvements in accuracy and F1 scores compared to existing methods. The results suggest that EF-Net has potential for real-world applications in mental state recognition and brain activity learning.