Noise Avoidance SMOTE in Ensemble Learning for Imbalanced Data

Class imbalance is a common problem in many real-world applications. To deal with class imbalance, several techniques, including resampling and ensemble approaches, have been proposed and resampling and ensemble methods have been proven effective for imbalanced data. Moreover, hybrid methods that co...

Full description

Saved in:
Bibliographic Details
Published in:IEEE access 2021, Vol.9, p.143250-143265
Main Author: Kim, Kyoungok
Format: Article
Language:English
Subjects:
Algorithms
Bagging
Boosting
Class imbalance
classification complexity
Complexity theory
Costs
Ensemble learning
Noise
Noise generation
Noise measurement
oversampling
Resampling
Sampling methods
SMOTE
Training
Citations: Items that this one cites
Items that cite this one
Online Access:Get full text
Tags: Add Tag
No Tags, Be the first to tag this record!
Description
Summary:Class imbalance is a common problem in many real-world applications. To deal with class imbalance, several techniques, including resampling and ensemble approaches, have been proposed and resampling and ensemble methods have been proven effective for imbalanced data. Moreover, hybrid methods that combine resampling and ensemble have been verified to be highly effective in dealing with imbalance problems. this study proposes new hybrid sampling/ensemble algorithms based on a modification of SMOTE, called NASBoost and NASBagging, which avoids selecting noise samples in the minority class while maintaining diversity among training sets. The proposed sampling method introduces new measures to identify samples that may generate noisy synthetic samples during sampling in SMOTE. Experimental results on 16 imbalanced datasets show that the hybrid of the proposed sampling procedure and ensemble algorithms improves the classification performance by preventing the generation of noise while allowing samples in the minority class to be evenly chosen.
ISSN:2169-3536
2169-3536
DOI:10.1109/ACCESS.2021.3120738