CHALLENGES OF AI-BASED TRAFFIC CLASSIFICATION IN MODERN TELECOMMUNICATION NETWORKS

Abstract

The increasing use of encrypted protocols and dynamic applications in modern telecommunication networks has significantly reduced the effectiveness of traditional traffic classification methods. Although artificial intelligence and machine learning algorithms can achieve high classification accuracy using flow-based features, accuracy alone cannot fully guarantee reliable performance under real-world network conditions. This study analyzes the key challenges associated with the practical deployment of AI-based traffic classifiers. In particular, it considers dynamic changes in network traffic, data imbalance, computational complexity, and the limited interpretability of model decisions. The analysis shows that traffic classifiers for real telecommunication networks should be evaluated using a multi-criteria approach that considers not only classification accuracy but also model generalization, resource efficiency, and adaptability.