Considerations of Using the Network Traffic Dataset for Machine Learning Algorithms

Recent advancements in network security have increasingly relied on machine learning (ML) algorithms to extract and analyze features from network traffic. These features are crucial for detecting malicious activities, identifying network services and classifying applications. However, the datasets used for training these ML models are often generated in controlled environments, leading to significant discrepancies between the characteristics of these datasets and those of actual network traffic. Researchers sometimes develop models that show significant performance on a given network traffic dataset by utilizing data from areas that should not be used. In this paper, we discuss the limitations of current network traffic datasets and emphasizes the need for realistic data representation to enhance the efficacy of ML-based network security solutions.