|
| << vorige | volgende >> |
Tijdschrift beschrijving
Alle jaargangen van het bijbehorende tijdschrift
Alle afleveringen van het bijbehorende jaargang
Alle artikelen van de bijbehorende aflevering
Details van artikel 6 van 11 gevonden artikelen
| Titel: |
Estimating harmfulness of class imbalance by scatter matrix based class separability measure |
| Auteur: |
Yu, Hualong Ni, Jun Xu, Sen Qin, Bin Jv, Hengrong |
| Verschenen in: |
Intelligent data analysis |
| Paginering: | Jaargang 18 (2014) nr. 2 pagina's 203-216 |
| Jaar: |
2014-03-04 |
| Inhoud: |
In many real world applications, class imbalance problems occur frequently, causing great underestimation for the classification performance of minority classes. In recent years, much effective solutions have been proposed to address this problem. However, the recent research found that not all skewed classification tasks are harmful and carrying out class imbalance learning methods on those unharmful ones can hardly improve and even degenerate classification performance, meanwhile increase training time to a large extent. Therefore, it is essential to design an efficient criterion to pre-estimate the harmfulness of class imbalance when we encounter skewed classification tasks. In this study, we explore the reason of harmfulness produced by class imbalance and propose a simple and ingenious strategy using scatter matrix based class separability measure to estimate the harmfulness of class imbalance with merely using training samples. The experimental results indicate that the proposed strategy can quantificationally estimate the damage for imbalanced classification tasks and provide priori information to guide us to select appropriate classification methods. Moreover, the computing complexity of the proposed strategy is quite low, thus it is practical in real-world applications. |
| Uitgever: |
IOS Press |
| Bronbestand: |
Elektronische Wetenschappelijke Tijdschriften |
Details van artikel 6 van 11 gevonden artikelen
| << vorige | volgende >> |