.png){kind=icon}
Return
Graph-Based Mutually Supervised Heterogeneous Ensemble Learning for Class-Imbalance Data
.png){kind=icon}
PRE
AI
DOI：10.1109/tsmc.2026.3713998.png)
.png){kind=icon}
Original request for help.png){kind=icon}
Share.png){kind=icon}
Save
Abstract
En 中文
Traditional homogeneous ensemble learning uses the same base classifier to improve the performance of the model by perturbing the training dataset or changing the instance weights, while its counterpart, the heterogeneous ensemble learning method using different base classifiers, is another way to improve the diversity of the model. However, how to correctly select base classifiers with complementary and diverse performance is a challenge faced by heterogeneous ensemble learning. Inspired by feature-enhanced stacking, we propose a graph-based mutually supervised heterogeneous ensemble learning (GMSHE) model. This method uses the idea of mutually supervised learning to balance the training data via different data augmentation methods on the data subsets obtained using bagging. Then, feature-enhanced stacking is used to evaluate the influence between models and establishes a global directed graph of the pool of base classifiers. Finally, a roulette wheel is used to search for the best base classifier combination in the global directed graph, and the results of the base classifiers are integrated using voting. Furthermore, during graph construction, the optimal classifier combination is determined according to the performance variation across different evaluation models, which helps reduce the risk of overfitting. Experimental results on 20 datasets indicate that, compared with the other seven ensemble learning methods, GMSHE performs better than other existing methods.
Keywords:
Ensemble learning
feature-enhanced stacking
heterogeneous ensemble learning
imbalanced data
mutually supervised learning
Journal
I
IF:
8.7
Papers:
142
Citations:
0
Organization
Cited Papers
No cited papers available