chYassine/LogAtlas-Defense-Set

LogAtlas-Defense-Set是一个异构的、带标签的日志数据集，旨在训练和评估日志级别和会话级别的分类器，以区分正常行为和跨多个来源（系统、网络和应用程序日志）的网络攻击。数据集的目标是作为LogAtlas生态系统的“防御层”，专注于在不同类别分布下进行稳健、现实的攻击检测。数据集包含多源日志（如系统、网络、应用程序日志），组织成带有丰富元数据的会话，包括攻击比例和严重性标签。适用于多种用例，如基于LLM或transformer的检测器训练、不同类别分布下的攻击检测基准测试以及严重性或源感知的威胁建模研究。

LogAtlas-Defense-Set is a heterogeneous, labeled log dataset designed for training and evaluating log-level and session-level classifiers that distinguish between normal behavior and cyberattacks across multiple sources (system, network, and application logs). It is intended as the “defense layer” of the LogAtlas ecosystem, focusing on robust, realistic attack detection under varied class distributions. The dataset includes multi-source logs (e.g., system, network, application) organized into sessions with rich metadata, including attack ratios and severity labels. It is suitable for various use cases such as training LLM-based or transformer-based detectors, benchmarking attack detection under different class distributions, and studying severity-aware or source-aware threat modeling.