The prevalence of cardiovascular diseases is increasing, of which stroke accounts for a large part. Stroke affects human cognitive or motor function, which is associated with quality of life after stroke, but studies predicting this prognosis in early stroke patients are still insufficient. In this study, the prognosis of motor function was predicted through the Fugl-Meyer Assessment (FMA), which is typical for measuring motor function in sub-acute stroke patients with difficulty in motor function. Specifically, the upper limb FMA score after 2 months was predicted as a more or less recovery group. Twelve sub-acute stroke patients participated, and electroencephalography (EEG) was measured for 5 min with their eyes closed. Features extracted mu and beta bands for frequency information and seven channels for spatial information associated with motor function. The model utilized EEGNet, which is most commonly used in EEG signals. As a result of the leave-one-subject-out cross-validation, the accuracy was 0.88±0.16. This performance outperformed the comparative model. This result suggests that focusing on motor function-related frequencies and channels contributes to enhanced prediction performance. Our study could be helpful for the allocation and planning of rehabilitative treatments at the individual level.