Existing public eye-tracking datasets rarely include data from deaf individuals, leaving a critical gap in related research. To address this, we constructed the Multimodal Deaf Eye Tracking Dataset (MDETD), which incorporates both spoken and signed language modalities. The dataset includes eye-tracking data from 27 deaf and 34 hearing participants, comprising approximately 1.47 million samples annotated into four categories: fixation, saccade, smooth pursuit, and noise. After data preprocessing and feature extraction, a 1D-CNN-BLSTM model was employed for eye movement classification. The model achieved F1 scores of 97.1% and 78.2% on fixation and saccade tasks, respectively, demonstrating strong classification performance. Further analysis revealed that deaf participants exhibited a significantly higher proportion of fixation behaviors, indicating enhanced visual concentration, while hearing participants showed more frequent saccades, suggesting differences in language modality usage and visual processing strategies. This work contributes empirical evidence to the understanding of visual cognition in the deaf population and provides a valuable resource for eye movement classification and cross-modal language processing research.