Skip to main content
stat
Imbalanced Classification based on a Geometric Graph Family
Add to Calendar 2021-03-19T14:10:00 2021-03-19T15:00:00 UTC Imbalanced Classification based on a Geometric Graph Family
Start DateFri, Mar 19, 2021
10:10 AM
to
End DateFri, Mar 19, 2021
11:00 AM
Presented By
Elvan Ceyhan (Auburn University)
Event Series: SMAC Talks

We use a geometric directed graph (digraph) family called class cover catch digraphs (CCCDs) for classification of imbalanced data. CCCDs are constructed based on spherical regions --- called proximity regions that determine the presence and direction of the arcs in the digraph--- and emerged as a graph theoretic solution to the class cover problem. We assess the classification performance of CCCD classifiers by extensive Monte Carlo simulations, comparing them with other classifiers commonly used in the literature. We show that CCCD classifiers perform well in an imbalanced classification setting where one class is more frequent than the other in a two-class setting. That is, CCCD classifiers are robust to the class imbalance problem in statistical learning. We also point out the relation between class imbalance and class overlapping problems, and their influence on CCCD classifiers and other classification methods including some of the state-of-the-art algorithms which are also robust to the class imbalance. CCCDs --- by construction --- tend to substantially under-sample from the majority class while preserving the information on the discarded points during the under-sampling process. While many state-of-the-art methods keep this information by means of ensemble classifiers, CCCDs yield only a single classifier with the same property, making it both simple and fast.