.png){kind=icon}
返回
Binary classification for imbalanced data using data conformity mechanism
.png){kind=icon}
PRE
AI
DOI：10.1007/s00530-024-01634-z.png)
.png){kind=icon}
原文求助.png){kind=icon}
分享.png){kind=icon}
收藏
摘要
En 中文
It is valuable to explore those hidden patterns from imbalanced data. In imbalanced data, skewed distribution of the classes makes minority classes to be hardly noticed. Existing classifiers easily suffer the perturbation caused by the skewed distribution so that they obtain unstable prediction and poor performance. Accordingly, to mitigate the perturbation, we utilize a score mechanism to employ a classifier being concerned about minority classes. Through calculating the conformity of the observed data, the score of the data conformity is obtained. And using the obtained score to classify highly imbalanced data. Following that, our classifier explores minority classes on the learning regions yielded by the score, instead of exploring them on the original data regions. Experimental results show that our classifier outperformed the competitors in classification performance and efficiency, moreover, it learned more compact boundaries separating minority classes from majority classes than the competitors did. Results also show that our classifier does not exhibit an exponential classification time at classifying large volume data with highly imbalanced ratio. We do not impose any restrictive classifier assumptions on both imbalance data and the calculation regarding the score of the data conformity. Additionally, we propose to sufficiently utilize the learning regions yielded by the score for better classification boundaries, instead of using the original data regions, since those hard-to-observe minority classes can be well perceived on the learning regions, meanwhile, there can tighten majority classes and minority classes so that the margins between them are significantly displayed.
Keyword:
Binary classification
Imbalance data
Score mechanism
期刊
IF:
3.1
论文数:
2.8K
被引数:
2.7K
机构
引用论文
.png){kind=icon}
Study on strength and durability characteristics of nano-silica based blended concrete纳米sio_2基混合混凝土强度及耐久特性研究
OAAI
OAAI分享收藏Error Analysis and Accuracy Calibration Method of U-Tube Coriolis Mass Flowmeter Under Pulsating Flow脉动流下u形管科氏质量流量计的误差分析及精度标定方法
PREAI
分享收藏Dynamically Weighted Balanced Loss: Class Imbalanced Learning and Confidence Calibration of Deep Neural Networks动态加权平衡损失: 深度神经网络的类不平衡学习和置信度校准
PREAI
分享收藏
查看更多内容