Добро пожаловать в клуб

Показать / Спрятать  Домой  Новости Статьи Файлы Форум Web ссылки F.A.Q. Логобург    Показать / Спрятать

Главное меню
ДомойНовостиСтатьиПостановка звуковФайлыДефектологияКнижный мирФорумСловарьРассылкаКаталог ссылокРейтинг пользователейЧаВо(FAQ)КонкурсWeb магазинШкольникамКарта сайта

Поздравляем нового Логобуржца Акулина со вступлением в клуб!



Handling Classification Uncertainty   Norbert Toth

Handling Classification Uncertainty

152 страниц. 2010 год.
LAP Lambert Academic Publishing
Many diagnostic applications require not only the samples to be classified, but a certainty value is required in addition to the predicted class label indicating the strength of the diagnosis. The first part of this Thesis proposes an extension to the decision tree framework to handle classification uncertainty, involving distance calculation to the relevant decision boundary; class density, correct classification probability and confidence estimation. The method is also applicable to trees that utilize oblique hyperplanes to cluster the input space and it is not restricted to the Euclidian distance metric. In the second part it is shown that these classification confidence values can be integrated to derive a consensus decision. Using the proposed combination scheme there is no need for an auxiliary combiner or weighting network, the weights are adaptively provided by the individual tree classifiers in the ensemble, new classifiers can be added dynamically without...
- Генерация страницы: 0.04 секунд -