Application-Level Anomaly Detection for the Master Caution Panel
Abstract
The goal of this work was to study how to monitor a large distributed system and apply machine learning methods to, and generate models of, its normal operation. With this done, the generated model(s) may be used to actively detect abnormal executions at run-time which may indicate improper use, attacks, or internal faults of the monitored system in question. We used the data collected by Master Caution Panel (MCP) software for the Theater Battle Management Core System (TBMCS) as sample data to test our machine learning methods. The MCP system has been under development for some time. It shares the same goal, but is based upon carefully designed and crafted "logic modules" that issue alerts when conditions warrant. The goal here is to use the existing monitoring and alert functions of MCP as a baseline to determine whether automated learning systems can achieve comparable performance in an automated fashion. A positive outcome of this study could suggest general principles of use in a wide range of mission critical systems.
Document Details
- Document Type
- Technical Report
- Publication Date
- Jul 01, 2005
- Accession Number
- ADA437103
Entities
People
- Salvatore J. Stolfo
Organizations
- Columbia University