Journal
NEURAL COMPUTING & APPLICATIONS
Volume 34, Issue 22, Pages 19923-19936Publisher
SPRINGER LONDON LTD
DOI: 10.1007/s00521-022-07534-5
Keywords
Broad learning system; Imbalance learning; Kernel learning; Binary classification
Categories
Funding
- Key-Area Research and Development Program of Guangdong Province [2019B010153002]
- National Natural Science Foundation of China [62106224]
- Marine Economy Development (Six Marine Industries) Special Foundation of Department of Natural Resources of Guangdong Province [GDNRC [2020]056]
Ask authors/readers for more resources
Broad Learning System (BLS) is a fast learning neural network that has shown good performance in various applications. However, conventional BLS has limitations in dealing with class imbalance and parameter tuning. To overcome these challenges, we propose a double-kernelized weighted broad learning system (DKWBLS) that improves feature representation and addresses class imbalance. Experimental results demonstrate the superiority of DKWBLS in handling imbalanced data.
Broad learning system (BLS) is an emerging neural network with fast learning capability, which has achieved good performance in various applications. Conventional BLS does not effectively consider the problems of class imbalance. Moreover, parameter tuning in BLS requires much effort. To address the challenges mentioned above, we propose a double-kernelized weighted broad learning system (DKWBLS) to cope with imbalanced data classification. The double-kernel mapping strategy is designed to replace the random mapping mechanism in BLS, resulting in more robust features while avoiding the step of adjusting the number of nodes. Furthermore, DKWBLS considers the imbalance problem and achieves more explicit decision boundaries. Numerous experimental results show the superiority of DKWBLS in tackling imbalance problems over other imbalance learning approaches.
Authors
I am an author on this paper
Click your name to claim this paper and add it to your profile.
Reviews
Recommended
No Data Available