Cost-sensitive budget adaptive label thresholding algorithms for large-scale online multi-label classification
by Rui Ding; Tingting Zhai
International Journal of Computational Science and Engineering (IJCSE), Vol. 27, No. 5, 2024

Abstract: Kernel-based methods have proven effective in addressing nonlinear online multi-label classification tasks. However, their scalability is hampered by the curse of kernelisation when handling large-scale tasks. Additionally, the class-imbalance of multi-label data can significantly impact their performance. To mitigate both challenges, we propose two cost-sensitive budget adaptive label thresholding algorithms. Firstly, we introduce a cost-sensitive strategy to assign varying costs to the misclassification of different labels, building upon the first-order adaptive label thresholding algorithm. Furthermore, we present two merging budget maintenance strategies: 1) a global strategy where all predictive models share one support vector pool and undergo simultaneous budgeting; 2) a separate strategy that utilises two independent support vector pools - one for the scoring models and the other for the thresholding model - with both models being budgeted independently. Experiments on seven datasets from various fields confirm the effectiveness and superiority of our proposed algorithms, particularly in large-scale online multi-label classification tasks.

Online publication date: Mon, 09-Sep-2024

The full text of this article is only available to individual subscribers or to users at subscribing institutions.

 
Existing subscribers:
Go to Inderscience Online Journals to access the Full Text of this article.

Pay per view:
If you are not a subscriber and you just want to read the full contents of this article, buy online access here.

Complimentary Subscribers, Editors or Members of the Editorial Board of the International Journal of Computational Science and Engineering (IJCSE):
Login with your Inderscience username and password:

Username:        Password:          Forgotten your password?

Want to subscribe?
A subscription gives you complete access to all articles in the current issue, as well as to all articles in the previous three years (where applicable). See our Orders page to subscribe.

If you still need assistance, please email subs@inderscience.com