Mosquito-borne diseases such as dengue (often transmitted by Aedes species), malaria (by Anopheles species), and West Nile Virus or lymphatic filariasis (by Culex species) pose persistent global health challenges, with millions of cases reported annually. Accurate identification of mosquito species is critical for early detection and targeted vector control, yet traditional methods remain labor-intensive and prone to errors. In this study, we propose a robust hybrid deep learning framework that integrates a pretrained Vision Transformer (ViT) for feature extraction with a Gaussian Naive Bayes classifier. The classifier is further optimized using the Virus Colony Search (VCS) algorithm to enhance generalization and stability. The model is evaluated on a balanced dataset containing 6000 high-resolution images across three key vector species: Aedes aegypti, Anopheles stephensi, and Culex quinquefasciatus. The proposed model achieved outstanding performance, with a test accuracy of 98.31%, precision of 98.32%, recall of 98.31%, and an F1-score of 98.31%, alongside high MCC (0.9734), CSI (0.9712), and FMI (0.9688). Compared to traditional CNN models and other ViT-ML combinations, the proposed approach demonstrated superior performance and generalization, particularly under class balance and domain variability. This framework presents a scalable and efficient solution for automated mosquito species classification and can be extended to incorporate additional modalities such as wingbeat audio or environmental data.