Solution Methods for Classification Problems with Categorical Attributes

• Published in: Computational mathematics and modeling Vol. 26; no. 3; pp. 408 - 428
• Main Author: D’yakonov, A. G.
• Format: Journal Article
• Language: English
• Published: New York Springer US 01.07.2015
• Subjects: Applications of Mathematics; Computational Mathematics and Numerical Analysis; Mathematical Modeling and Industrial Mathematics; Mathematics; Mathematics and Statistics; Optimization; categorical attribute; factor attribute; singular decomposition; nominal attribute; classification; category; factor; machine learning; encoding
• ISSN: 1046-283X; 1573-837X
• DOI: 10.1007/s10598-015-9281-2

The article considers various methods for classification of a set of objects into two classes when all the attributes are categorical (nominal or factor attributes), i.e., describe the membership of an object in a category. Some methods are a simple generalization of classical methods (Bayesian algorithms, singular decomposition methods), others are fundamentally novel. An efficient technique is proposed for encoding categorical attributes by real numbers, which makes it possible to apply classical machine-learning methods (e.g., the random forest). A generalization of the k nearest neighbors (kNN) algorithm and Zhuravlev’s estimate calculation algorithm (AEC) achieve best performance on real-life data. All methods have been tested on an applied problem involving construction of a recommender system for a security service.