A Logical and Probabilistic Technique for Classification and Dimensionality Reduction for Objects with Categorical Data
Abstract
A supervised learning technique, the Attribute Importance Measure (AIM) method, is proposed for the classification of objects with categorical attributes. The advantage of this method over existing techniques is its ability to perform classification and dimensionality reduction, or feature selection, with the same algorithm. The method uses probabilistic measures alongside logical concepts of sufficiency, necessity and irrelevance in providing corresponding weights to values in attribute value pairs. Finally an efficient search algorithm is developed which generates decision rules for classification. The performance of the new method is demonstrated on a commonly used machine learning data set.
Document Details
- Document Type
- Technical Report
- Publication Date
- Jun 01, 2004
- Accession Number
- ADA427562
Entities
People
- Mark Porter
Organizations
- Defence Science and Technology Group