With the widespread adoption of Content Delivery Networks (CDNs), the issue of imbalanced traffic samples has become increasingly prominent in the task of CDN brand identification by encrypted traffic. To address this challenge, this paper proposes a novel method, Classifier Evolutionary Generative Adversarial Network (CIEGAN). By generating traffic data for underrepresented categories, CIEGAN alleviates the recognition difficulties caused by data imbalance. It incorporates the concept of hierarchical evolutionary algorithms to expand the search space and enhance generation capabilities. Additionally, a classifier-guided mechanism is introduced during adversarial training. Experimental results demonstrate that CIEGAN outperforms traditional methods and other GAN-based models. Its NLL loss is over 0.1 lower than the second best method. The generated data exhibit visual distributions closely aligned with real data. Furthermore, CIEGAN significantly improves the classification accuracy of these categories, achieving a 12% improvement in the F1-score compared to the baseline. Importantly, this enhancement is achieved without compromising the classification performance of other categories, leading to an overall F1-score improvement of more than 1%.