Apportioned Margin Approach for Cost Sensitive Large Margin Classifiers
This work addresses the problem of prioritizing important classes in classification for applications like medical diagnosis or fraud detection, though it appears incremental as it builds on existing margin-based methods.
The paper tackles cost-sensitive multiclass classification by introducing an apportioned margin framework that shifts margins between classes based on prioritization, resulting in a tighter error bound for important classes and reduced overall out-of-sample error, with promising empirical results.
We consider the problem of cost sensitive multiclass classification, where we would like to increase the sensitivity of an important class at the expense of a less important one. We adopt an {\em apportioned margin} framework to address this problem, which enables an efficient margin shift between classes that share the same boundary. The decision boundary between all pairs of classes divides the margin between them in accordance to a given prioritization vector, which yields a tighter error bound for the important classes while also reducing the overall out-of-sample error. In addition to demonstrating an efficient implementation of our framework, we derive generalization bounds, demonstrate Fisher consistency, adapt the framework to Mercer's kernel and to neural networks, and report promising empirical results on all accounts.