Full Citation

Title: Large Scale Empirical Comparison of Linear Classifiers for Multi-class Problems

Citation Type: Miscellaneous

Publication Year: 2014

ISBN:

ISSN:

DOI:

NSFID:

PMCID:

PMID:

Abstract: Linear classifiers, even though very simple, are pop- ular for classification tasks. By nature they can only differentiate between two classes. But it is possible to extend their usage into the domain of multi-class problems. These classifiers are known to perfrom very well for many practical scenarios (both binary and multi-class). In this paper, we examine and study the performance of linear classifiers across different data sets for multi-class clas- sification. Specifically, we compare the performance of Logistic Regression, Naive Bayes, Linear SVM, Weighted Majority and ECOC (Error Correcting Output Codes) constructed using Naive Bayes across various data sets. In addition to that, we show how these classifiers perform when the data sets have only numeric attributes, only nominal attributes and mixed attributes. We also study about the performance of these linear classifiers on large data sets and class imbalance problems. Our study shows that different linear classifiers perform better in different situations.

Url: http://pages.cs.wisc.edu/~anubhavnidhi/large-scale-empirical.pdf

User Submitted?: No

Authors: Sen, Ayon; Narayanaswamy, Ashwin, K; Abhashkumar, Anubhavnidhi

Publisher: University of Wisconsin

Data Collections: IPUMS USA

Topics: Population Data Science

Countries: United States