A Data Mining approach for building cost-sensitive and light intrusion detection models
Abstract
The report provides a summary of the intrusion detection research completed for this effort. The research studied how to build cost-sensitive and light weight intrusion detection models. The main technical components of the research are: 1) Automatic feature construction by analyzing the patterns of normal and intrusion activities computed from large amounts of audit data. 2) Using cost-sensitive machine learning algorithms to construct intrusion detection models that achieve optimal performance on the given (often site-specific) cost metrics, cluster attack signatures and normal profiles and accordingly construct one light model of each cluster to maximize the utility of each model. 3) Dynamic (re-) configuration of the light models to make an IDS effective and efficient, and resilient to IDS-related attacks. Algorithms and prototype systems were developed and extensive experiment using DARPA datasets and other real-world datasets were conducted. The results showed that the technologies developed in this project are more advanced and better than today's state-of-the-art.
Document Details
- Document Type
- Technical Report
- Publication Date
- Mar 01, 2004
- Accession Number
- ADA422555
Entities
Organizations
- North Carolina State University