Novel DDoS Detection Dataset for Cross-Domain Training

This technical report introduces SCLDDOS2024, a collection of four diverse datasets aimed at enhancing DDoS detection research. The datasets were gathered over a period of 21 months from various commercial networks using consistent methods and tools. The report outlines the limitations of existing datasets, which are primarily synthetic or limited in scope, and emphasizes the need for more robust data to improve model generalization. The datasets are collected using commercially available DDoS protection solutions, although they may not be optimal for all research purposes due to constraints in real-world data collection. The report details how even basic machine learning models can achieve high accuracy when paired with effective feature engineering applied to SCLDDOS2024. Additionally, the report discusses the unique characteristics of the datasets, including their ability to capture a wide range of attack patterns and anomalies, which are often missing in synthetic datasets. This work aims to provide richer, more diverse data for the DDoS detection research community.