.png){kind=icon}
.png){kind=icon}
Return
A multi-layer Bayesian trial-and-error learning algorithm for imbalanced classification
.png){kind=icon}
PRE
AI
DOI：10.1016/j.engappai.2026.114444.png)
.png){kind=icon}
Original request for help.png){kind=icon}
Share.png){kind=icon}
Save
Abstract
En 中文
Feature optimization, sampling strategies, and classifiers are critical for enhancing the recognition of minority classes in imbalanced classification. However, existing approaches often select these techniques separately, leading to inadequate combinations and thereby failing to adapt effectively across multiple imbalanced datasets. Thus, we propose a multi-layer Bayesian trial-and-error learning algorithm (MBTE) that facilitates collaborative selection and combination optimization. The working principle at the heart of MBTE is a Bayesian multi-layer decision-making scheme for collaborative selection (BMDC), where the selection of each layer depends on the outcome of the previous layer. To assess the combination selected by BMDC, a weighted multi-metric-based reward function scheme (WMRS) is used to quantify the rewards of the combination based on multiple weighted metrics. Furthermore, a Thompson sampling-based scheme for optimal exploration strategy (TSOE) explores the diversity of the combination, enabling the selection strategy to be optimized. With these three schemes, MBTE is made capable of optimizing the decision-making process and selecting the optimal combination for imbalanced classification. Finally, experimental results show that MBTE outperforms state-of-the-art algorithms via 12 public imbalanced datasets in terms of three widely used evaluation metrics for classification. To further validate the performance of MBTE, we also discuss the impact of the multi-layer structure order, reward function, and exploration strategy. The results demonstrate the effectiveness of the MBTE, which could be extended to some applications of artificial intelligence (AI), such as disease diagnosis and event detection.
Keywords:
imbalanced classification
Bayesian optimization
feature optimization
sampling strategies
classifier selection
Journal
IF:
8
Papers:
5.3K
Citations:
3.5W