NetBench: A Large-Scale and Comprehensive Network Traffic Benchmark Dataset for Foundation Models

在计算机网络中，网络流量指的是在互联的计算机或系统之间以数据包形式传输的数据量。监控和分析网络流量对于确保网络的性能、安全性和可靠性至关重要。然而，网络流量分析中的一个重要挑战是处理包括密文和明文在内的各种数据包。虽然已经采用了许多方法来分析网络流量，但它们通常依赖于不同的数据集进行性能评估。这种不一致性导致了大量的手动数据处理工作和不公平的比较。此外，一些数据处理方法可能会由于训练和测试数据的不当分离而导致数据泄漏。为了解决这些问题，我们介绍了NetBench，这是一个用于评估机器学习模型（特别是基础模型）在流量分类和生成任务中的大规模和全面的基准数据集。NetBench基于七个公开可用的数据集，包括15个分类任务和5个生成任务的广泛范围的20个任务。此外，我们使用我们的基准测试评估了八个最先进的分类模型和两个生成模型。结果表明，基础模型在流量分类方面明显优于传统的深度学习方法。我们相信NetBench将有助于各种方法之间的公平比较，并推进基础模型在网络流量方面的发展。我们的基准数据集可在 https://github.com/WM-JayLab/NetBench 上获取。

NetBench: A Comprehensive Benchmark for Evaluating Machine Learning Models in Network Traffic Classification and Generation

The paper presents NetBench, a large-scale and comprehensive benchmark dataset for assessing machine learning models in both traffic classification and generation tasks. The dataset is built upon seven publicly available datasets and encompasses a broad spectrum of 20 tasks. The paper also evaluates eight State-Of-The-Art (SOTA) classification models and two generative models using the benchmark.

The experiments show that foundation models significantly outperform traditional deep learning methods in traffic classification. NetBench will facilitate fair comparisons among various approaches and advance the development of foundation models for network traffic. The benchmark is available at https://github.com/WM-JayLab/NetBench.

Some related studies in this field include 'Deep Packet: A Novel Approach for Encrypted Traffic Identification Using Deep Learning' and 'A Survey on Deep Learning for Network Anomaly Detection'.