Sleep arousal detection plays a pivotal role in diagnosing and managing sleep disorders. The paper introduces a unique framework for the automated detection for sleep arousal events, leveraging a multi-modal convolutional neural network (CNN) architecture enhanced with attention mechanisms. The proposed method employs the Sleep Heart Health Study (SHHS) dataset, which provides comprehensive polysomnographic (PSG) recordings encompassing EEG, EOG, and EMG signals. The approach is designed to capture the diverse and complementary physiological information inherent in these signals by processing 30-s epochs of PSG data through distinct convolutional pathways for each modality. The subsequent feature fusion and attention weighting stages are tailored to selectively emphasize the most relevant features selectively, optimizing the model’s focus and performance. Comprehensive evaluations of the SHHS dataset reveal the efficacy of the proposed method, with the model achieving a notable accuracy of 95.16% and an Area Under the Receiver Operating Characteristic Curve (AUROC) score with 0.94. These results significantly advance over conventional techniques and single-modality CNN architectures, underscoring the advantages of a multi-modal approach in sleep arousal detection. The integration of attention mechanisms further enhances the model’s ability to address the complexities associated with manual annotation, providing a scalable and efficient solution for clinical application. This work contributes to the field of sleep disorder diagnosis and offers a robust tool for clinicians, facilitating the accurate detection of sleep arousal events. Future research will focus on incorporating additional physiological signals and validating the model’s performance in real-time clinical settings, aiming to extend its applicability and clinical utility further.