A Comprehensive Approach To Outlier Detection and Event Classification.
Abstract
In this report, a comprehensive approach to outlier detection and event classification is investigated. Although the methodology is based on the assumption of available training data, it does not require ground truth or labels of any type. In fact, it is not even required that the number of different populations composing the training data is known. Data from Western China is analyzed to demonstrate the methodology, as well as some simulated data. These examples demonstrate vividly the importance of the role of correlation in selecting the best features. A method for feature selection is considered. Additionally, the problems of classifying events, numerical stability, missing data, signal to noise ratios, and mixture (discrete and continuous) data are discussed.
Document Details
- Document Type
- Technical Report
- Publication Date
- Dec 01, 1996
- Accession Number
- ADA324277
Entities
People
- H. L. Gray
- S. R. Sain
- W. A. Woodward
Organizations
- Southern Methodist University