Learning From Imbalanced Data
By Lincy Mathews, Seetha Hari. 2019
4,781 citations recorded by the source corpus.
A very challenging issue in real-world data is that in many domains like medicine, finance, marketing, web, telecommunication, management, etc. the distribution of data among classes is inherently imbalanced. A widely accepted researched issue is that the traditional classifier algorithms assume a balanced distribution among the classes. Data imbalance is evident when the number of instances representing the class of concern is much lesser than other classes. Hence, the classifiers tend to bias towards the well-represented class. This leads to a higher misclassification rate among the lesser represented class. Hence, there is a need of efficient learners to classify imbalanced data. This chapter aims to address the need, challenges, existing methods, and evaluation metrics identified when learning from imbalanced data sets. Future research challenges and directions are highlighted.
Subfields: Artificial Intelligence, Electrical and Electronic Engineering, Health Information Management, Representation & Self-Supervised Learning
How researchers see this paper
The researchers whose trusted networks rank it highest, strongest advocate first. Every researcher gets a different order.
In Artificial Intelligence
- Jiashi Feng
- Bernt Schiele
- Dacheng Tao
- Trevor Darrell
- Yoshua Bengio
- Jianfeng Gao
- Mohit Bansal
- Percy Liang
In Electrical and Electronic Engineering
- Francisco Herrera
- Gianluca Bontempi
- Andrea Dal Pozzolo
- Alberto Fernández
- Sriraam Natarajan
- Salvador Garcia
- Olivier Caelen
- Lichao Sun
In Health Information Management
- B. Van calster
- Dhiraj Neupane
- Mohammed Ali Al-Garadi
- Jun Shu
- Alberto Fernández
- Jun Yan
- Xiang Zhang
- Ewout W. Steyerberg
In Representation & Self-Supervised Learning
- Jiashi Feng
- Martial Hebert
- Shaogang Gong
- Yu-Xiong Wang
- Deva Ramanan
- Ziwei Liu
- Xiatian Zhu
- Chen Change Loy
Citations in the corpus
- HateCheck: Functional Tests for Hate Speech Detection Models (2020, background): “use class weights emphasising the hateful minority class (He and Garcia, 2009).”
- A Survey on Data Collection for Machine Learning: A Big Data - AI Integration Perspective (2018, background): “He and Garcia [133] provide a comprehensive survey on learning from imbalanced data.”
- Using Pre-Training Can Improve Model Robustness and Uncertainty (2019, background): “To this end, He & Garcia (2008) propose to remove samples from the majority classes, while Huang et al. (2016) replicate samples from the minority classes.”
- IL2M: Class Incremental Learning With Dual Memory (2019, background): “Here, the class IL problem becomes akin to an imbalanced learning one [9].”
- The Devil is in the Tails: Fine-grained Classification in the Wild (2017, methodology): “In the first regime, cost-sensitive learning [10] is employed to force the model to adjust its decision boundaries by incurring a non-uniform cost per misclassification; see [19] for a review of the techniques.”
- Class-Balanced Loss Based on Effective Number of Samples (2019, neutral): “CNNs trained on such data perform poorly for weakly represented classes [19, 15, 41, 4].”
- Generalizing from a Few Examples (2019, background): “• Imbalanced learning [49] learns from experience E with a severely skewed distribution for y.”
- Tencent ML-Images: A Large-Scale Multi-Label Image Database for Visual Representation Learning (2019, background): “The cost parameter η > 1 is thus introduced to set a larger cost on positive labels than negative labels, to alleviate the imbalance between positive and negative images in each category, which is a common strategy in imbalance learning [29].”
- Unbiased Scene Graph Generation From Biased Training (2020, neutral): “With the power of counterfactual causality, we can remove the harmful bias from the good context bias, which cannot be easily identified by traditional debiasing methods such as data augmentation [9, 11] and unbiased learning [30].”
- An up-to-date comparison of state-of-the-art classification algorithms (2017, background): “For example, some measures are inappropriate for imbalanced data or multiclass learning, while others can reveal more information in such cases ( Caruana & Niculescu-Mizil, 2006; He & Garcia, 2009 ).”
- Multiscale Convolutional Neural Networks for Fault Diagnosis of Wind Turbine Gearbox (2019, methodology): “formance evaluation and comparison, F1 score [37] is adopted in this study.”
- Comparison of Random Forest, k-Nearest Neighbor, and Support Vector Machine Classifiers for Land Cover Classification Using Sentinel-2 Imagery (2017, background): “He and Garcia [56] stated that if input datasets (training samples) are highly imbalanced, the OA value might be deceiving, because the rare classes may be classified very poorly.”
- Combination of Domain Knowledge and Deep Learning for Sentiment Analysis (2017, methodology): “Since our data is unbalanced between positive, neutral and neutral samples, we use the SMOTE [23] sampling method to balance data.”
- Queried Unlabeled Data Improves and Robustifies Class-Incremental Learning (2022, positive): “Many of those algorithms viewed CIL as an imbalanced learning problem, where previous and newly added classes made an class size extreme disparity (He & Garcia, 2009; Buda et al., 2018).”
- Structure learning for relational logistic regression: an ensemble approach (2018, background): “The arguments follow (Yang et al. 2014), and we pick our λ from the specific range using domain knowledge.”
- Class-Weighted Classification: Trade-offs and Robust Approaches (2020, background): “We briefly review other research related to imbalanced classification, but for a far more exhaustive treatment, see a survey of the area (He & Garcia, 2009; Fernández et al., 2018).”
- A Deep Learning Architecture for Temporal Sleep Stage Classification Using Multivariate and Multimodal Time Series (2017, background): “One way to address this issue is to reweight the model loss function so that the cost of making an error on a rare sample is larger [15].”
- Comprehensive Review of Artificial Neural Network Applications to Pattern Recognition (2019, methodology): “GAN is applying to the future prediction of the frame and in high dimensional spaces such as 3D modeling [112].”
- Incremental Event Detection via Knowledge Consolidation Networks (2020, background): “The ambiguous examples could confuse the classifier, affecting the generalization ability (Liu et al., 2014 (He and Garcia, 2009; Wu et al., 2019), which severely misleads the classifier.”
- Machine learning for the prediction of sepsis: a systematic review and meta-analysis of diagnostic test accuracy (2020, background): “First, the AUROC was pragmatically chosen as a summary measure, while it may underperform in the setting of imbalanced datasets [60].”