This paper surveys techniques for classifying Internet traffic using Machine Learning (ML), focusing on methods that do not rely on well-known TCP/UDP port numbers or payload inspection. It reviews 18 significant works from 2004 to early 2007, categorizing them based on ML strategies and contributions. The paper discusses key requirements for ML-based traffic classifiers in operational networks, critiques the reviewed works, and identifies open issues and challenges in the field. It also explores the importance of IP traffic classification in network management, lawful interception, and QoS. The paper highlights the limitations of traditional packet inspection methods, such as port-based and payload-based classification, and introduces newer approaches that use statistical traffic properties for classification. It discusses supervised and unsupervised learning techniques, feature selection, and challenges in operational deployment, including timely classification, directional neutrality, memory and processor efficiency, and robustness. The paper concludes with a review of ML-based IP traffic classification techniques, categorizing them into clustering, supervised learning, hybrid, and comparison approaches.This paper surveys techniques for classifying Internet traffic using Machine Learning (ML), focusing on methods that do not rely on well-known TCP/UDP port numbers or payload inspection. It reviews 18 significant works from 2004 to early 2007, categorizing them based on ML strategies and contributions. The paper discusses key requirements for ML-based traffic classifiers in operational networks, critiques the reviewed works, and identifies open issues and challenges in the field. It also explores the importance of IP traffic classification in network management, lawful interception, and QoS. The paper highlights the limitations of traditional packet inspection methods, such as port-based and payload-based classification, and introduces newer approaches that use statistical traffic properties for classification. It discusses supervised and unsupervised learning techniques, feature selection, and challenges in operational deployment, including timely classification, directional neutrality, memory and processor efficiency, and robustness. The paper concludes with a review of ML-based IP traffic classification techniques, categorizing them into clustering, supervised learning, hybrid, and comparison approaches.