Anzeige
Erschienen in:
2018 | OriginalPaper | Buchkapitel
Improved Performance of EK-NNClus by Selecting Appropriate Parameter
verfasst von : Qian Wang, Zhi-gang Su
Erschienen in: Belief Functions: Theory and Applications
Aktivieren Sie unsere intelligente Suche, um passende Fachinhalte oder Patente zu finden.
Wählen Sie Textabschnitte aus um mit Künstlicher Intelligenz passenden Patente zu finden. powered by
Markieren Sie Textabschnitte, um KI-gestützt weitere passende Inhalte zu finden. powered by
Abstract
EK-NNclus is an evidential clustering method based on the evidential K-nearest neighbors classification rule. Its one significant merit is that it does not require any priori on the number of clusters. However, the EK-NNclus suffers from the influence of number K. In other words, the performance of EK-NNclus is sensitive to K: if the number K is too small, the natural cluster may be split into two or more clusters; otherwise, two or more natural clusters may be merged into one cluster. In this paper, we indicated that tuning the parameters (such as \(\alpha \) in the discounting function) can take full advantage of the distances between the object and its nearest neighbors, which can prevent natural clusters from being merged. Some numerical experiments were conducted and the experimental results suggested that the performance of EK-NNclus can be improved if appropriate \(\alpha \) is selected.