Impact of Classifier Prototyping on Classification Systems Optimization
Paulo V. W. Radtke, Robert Sabourin, Tony Wong · 2011
This paper compares two wrapper approaches to op- timize classification systems. The traditional direct ap- proach uses the actual classifier on the optimization process, whereas the second prototypes the target clas- sifier with a computationally inexpensive classifier. Our results indicate that using the actual classifier yields com- putationally faster classifiers. However, the accuracy in both approaches is comparable and the required optimiza- tion time may be prohibitive in certain contexts, justifying the use of a prototyping approach.