| Unbalanced datasets generate difficulties in designing good classification models because those classes that are represented by the most numerous training sets are harmfully preferred. For this reason, learning sets are often balanced by adding some synthetic feature vectors or by reducing the most numerous learning sets. High-dimensional learning sets give possibility to design complex layer of linear classifiers. Such layers can also be used for balancing purposes. In this approach, averaging of a small number of feature vectors is partially complemented by averaging vertices based on balanced feature subsets. |
*** Title, author list and abstract as seen in the Camera-Ready version of the paper that was provided to Conference Committee. Small changes that may have occurred during processing by Springer may not appear in this window.