The increasing popularity of wearable eye-Tracking systems has led to growing interest in monitoring and predicting mental disorders and dementia in mobile health (m-health) applications. However, collecting eye gaze data is a challenging and time-consuming process due to the heterogeneity of human visual behavior and privacy concerns. This scarcity of data limits the performance of classifiers. Even with data augmentation and deep generative models, the newly generated datasets may still suffer from the problem of domain shift, where the training and test data distributions do not match. To address this issue, this paper proposes a model-driven simulation approach for synthesizing eye gaze dynamics data based on clinical statistics during standard visual cognitive assessments. The synthesized dataset provides a more balanced and annotated set of signals that can enhance the performance gaze activity recognition.