Bayesian multiple imputation approaches for one-class classification

Authors:
Shehroz S. Khan;Jesse Hoey;Daniel Lizotte
Affiliations:
David R. Cheriton School of Computer Science, Univesity of Waterloo, Canada;David R. Cheriton School of Computer Science, Univesity of Waterloo, Canada;David R. Cheriton School of Computer Science, Univesity of Waterloo, Canada
Venue:
Canadian AI'12 Proceedings of the 25th Canadian conference on Advances in Artificial Intelligence
Year:
2012

Citing 5
Cited 1

Support Vector Data Description

Machine Learning
One-Class Classification by Combining Density and Class Probability Estimation

ECML PKDD '08 Proceedings of the 2008 European Conference on Machine Learning and Knowledge Discovery in Databases - Part I
Using Imputation Techniques to Help Learn Accurate Classifiers

ICTAI '08 Proceedings of the 2008 20th IEEE International Conference on Tools with Artificial Intelligence - Volume 01
The WEKA data mining software: an update

ACM SIGKDD Explorations Newsletter
LIBSVM: A library for support vector machines

ACM Transactions on Intelligent Systems and Technology (TIST)

Density weighted support vector data description

Expert Systems with Applications: An International Journal

Quantified Score

Hi-index	0.00

Visualization

Abstract

One-Class Classifiers build classification models in the absence of negative examples, which makes it harder to estimate the class boundary. The predictive accuracy of one-class classifiers can be exacerbated by the presence of missing data in the positive class. In this paper, we propose two approaches based on Bayesian Multiple Imputation (BMI) for imputing missing data in the one-class classification framework called Averaged BMI and Ensemble BMI. We test and compare our approaches against the common method of Mean imputation and Expectation Maximization on several datasets. Our preliminary experiments suggest that as the missingness in the data increases, our proposed imputation approaches can do better on some data sets.