Cited by 3,088Imbalanced Data Classification TechniquesAdvanced Neural Network ApplicationsMachine Learning and Data Classification

The DOI matches its Crossref record; 77 of the 83 references it cites were found in a registry or an index.

A systematic study of the class imbalance problem in convolutional neural networks

Open access

Buda, Mateusz, Maki, Atsuto, Mazurowski, Maciej A.·2018·Neural Networks

DOI 10.1016/j.neunet.2018.07.011

Open this paper in CoChat — free account Open PDF

Abstract

In this study, we systematically investigate the impact of class imbalance on classification performance of convolutional neural networks (CNNs) and compare frequently used methods to address the issue. Class imbalance is a common problem that has been comprehensively studied in classical machine learning, yet very limited systematic research is available in the context of deep learning. In our study, we use three benchmark datasets of increasing complexity, MNIST, CIFAR-10 and ImageNet, to investigate the effects of imbalance on classification and perform an extensive comparison of several methods to address the issue: oversampling, undersampling, two-phase training, and thresholding that compensates for prior class probabilities. Our main evaluation metric is area under the receiver operating characteristic curve (ROC AUC) adjusted to multi-class tasks since overall accuracy metric is associated with notable difficulties in the context of imbalanced data. Based on results from our experiments we conclude that (i) the effect of class imbalance on classification performance is detrimental; (ii) the method of addressing class imbalance that emerged as dominant in almost all analyzed scenarios was oversampling; (iii) oversampling should be applied to the level that completely eliminates the imbalance, whereas the optimal undersampling ratio depends on the extent of imbalance; (iv) as opposed to some classical machine learning models, oversampling does not cause overfitting of CNNs; (v) thresholding should be applied to compensate for prior class probabilities when overall number of properly classified cases is of interest.

Integrity signals

Registry

Crossref · 10.1016/j.neunet.2018.07.011

Cites

77 of 77 references verified0 citation errors · 0 not checkable

Statistics

Recomputing p-values needs the paper’s full text.

Claims the paper makes

We couldn’t retrieve this paper’s full text, so its claims and self-check weren’t run. The registry and citation checks still apply.

Claims other papers make about this one

Each is checked against this paper’s own text: does it say what the citing paper says it does?

Deep Residual Learning for Image Recognition
He, K., Zhang, X., Ren, S., & Sun, J. (2016) · 2016 IEEE Conference on Computer Vision and Pattern Recognition (CVPR) · doi:10.1109/cvpr.2016.90 · Cited by 235,474 · cited 1×
“And for all the cases, the classes chosen to be minority were the ones with the lowest label value.”
Background
Very Deep Convolutional Networks for Large-Scale Image Recognition
Simonyan, K., & Zisserman, A. (2014) · arXiv · doi:10.48550/arxiv.1409.1556 · Cited by 114,148 · cited 1×
“A standard CNN architecture comprises fully connected layers and a number of blocks consisting of convolution kernels, activation function layer and max pooling [22, 23, 24].”
Method
ImageNet classification with deep convolutional neural networks
Krizhevsky, A., Sutskever, I., & Hinton, G. E. (2017) · Communications of the ACM · doi:10.1145/3065386 · Cited by 112,155
Scikit-learn: Machine Learning in Python
Pedregosa, F., Varoquaux, G., Gramfort, A., Michel, V., Thirion, B., Grisel, O., Blondel, M., Müller, A., Nothman, J., Louppe, G., et al. (2012) · arXiv · doi:10.48550/arxiv.1201.0490 · Cited by 71,016 · cited 1×
“We used a speciﬁc implementation of the ROC AUC available in scikit-learn python package [65].”
Method
Gradient-based learning applied to document recognition
Lecun, Y., Bottou, L., Bengio, Y., & Haffner, P. (1998) · Proceedings of the IEEE · doi:10.1109/5.726791 · Cited by 58,805 · cited 2×
“Images The CNN model that we use for MNIST is the modern version of LeNet-5 [52].”
Method
ImageNet Large Scale Visual Recognition Challenge
Russakovsky, O., Deng, J., Su, H., Krause, J., Satheesh, S., Ma, S., Huang, Z., Karpathy, A., Khosla, A., Bernstein, M., et al. (2015) · International Journal of Computer Vision · doi:10.1007/s11263-015-0816-y · Cited by 39,380 · cited 1×
“In our study, we used three benchmark datasets: MNIST [52], CIFAR-10 [53] and ImageNet Large Scale Visual Recognition Challenge (ILSVRC) 2012 [54].”
Method
SMOTE synthetic minority over-sampling technique
Chawla, N. V., Bowyer, K. W., Hall, L. O., & Kegelmeyer, W. P. (2002) · Journal of Artificial Intelligence Research · doi:10.1613/jair.953 · Cited by 32,865 · cited 3×
“For classical machine learning models it was shown that oversampling can cause overﬁtting, especially for minority classes [33].”
BackgroundMethod
Delving Deep into Rectifiers: Surpassing Human-Level Performance on ImageNet Classification
He, K., Zhang, X., Ren, S., & Sun, J. (2015) · 2015 IEEE International Conference on Computer Vision (ICCV) · doi:10.1109/iccv.2015.123 · Cited by 19,741 · cited 1×
“Network weights were initialized with Kaiming (also known as MSRA) initialization procedure [63].”
Method
Visualizing and understanding convolutional networks
Zeiler, M. D., & Fergus, R. (2014) · Lecture Notes in Computer Science · doi:10.1007/978-3-319-10590-1_53 · Cited by 15,839 · cited 1×
“This property allows CNNs to learn hierarchical representations [2].”
Background
Batch Normalization: Accelerating Deep Network Training by Reducing Internal Covariate ShiftPreprint
Ioffe, S. (2024) · arXiv (Cornell University) · doi:10.57702/o9raffed · Cited by 15,273

Show all 83 references

Sorted by citations

Items 901–1,000 of 1,001

Automatic Identification of Vulnerable Code: Investigations with an AST-Based Neural Network
Partenza, G., Amburgey, T., Deng, L., Dehlinger, J., & Chakraborty, S. (2021) · IEEE Annual International Computer Software and Applications Conference (COMPSAC) · doi:10.1109/compsac51774.2021.00219 · Cited by 16
Data Augmentation for Imbalanced HRRP Recognition Using Deep Convolutional Generative Adversarial Network
Song, Y., Li, Y., Wang, Y., & Hu, C. (2020) · IEEE Access · doi:10.1109/access.2020.3032580 · Cited by 16
Multi-task Regularization Based on Infrequent Classes for Audio CaptioningPreprint
Çakır, E., Drossos, K., & Virtanen, T. I. (2020) · arXiv (Cornell University) · doi:10.48550/arxiv.2007.04660 · Cited by 15
“Though, in many natural language-based datasets (like Clotho) it is observed that there is a typical class imbalance [9, 10].”
Background
A Comparison of Synthetic Oversampling Methods for Multi-class Text ClassificationPreprint
Glazkova, A. V. (2020) · arXiv (Cornell University) · doi:10.48550/arxiv.2008.04636 · Cited by 16
“The minority class examples can be identiﬁed by neural network as noise, and therefore they could be wrongly discarded by the classiﬁer [9,10].”
Background
Photometric redshift estimation with convolutional neural networks and galaxy images: Case study of resolving biases in data-driven methods
Lin, Q., Fouchez, D., Pasquet, J., Treyer, M. A., Ouahmed, R. A., Arnouts, S., & Ilbert, O. (2022) · Astronomy and Astrophysics · doi:10.1051/0004-6361/202142751 · Cited by 16
Imbalanced Node Classification With Synthetic Over-Sampling
Zhao, T., Zhang, X. D., & Wang, S. (2024) · IEEE Transactions on Knowledge and Data Engineering · doi:10.1109/tkde.2024.3443160 · Cited by 19
Advancing Grapevine Variety Identification: A Systematic Review of Deep Learning and Machine Learning Approaches
Carneiro, G. A., Cunha, A. M. T. D. S., Aubry, T., & Sousa, J. J. (2024) · AgriEngineering · doi:10.3390/agriengineering6040277 · Cited by 16
A novel sparse-aware contrastive learning network with adaptive gating neurons for extreme class imbalance diagnosis scenarios
Guo, P., Huang, W., Ding, C., Shi, J., & Zhu, Z. (2025) · Mechanical Systems and Signal Processing · doi:10.1016/j.ymssp.2025.112895 · Cited by 17
A deep learning-based framework for large-scale plant disease detection using big data analytics in precision agriculture
Elfouly, M. K., AbdelAziz, A. M., Gomaa, W. H., & Abdalla, M. (2025) · Journal Of Big Data · doi:10.1186/s40537-025-01265-9 · Cited by 15
Machine learning prediction of hydrochar adsorption capacity for methylene blue with limited data: Inspired by generative adversarial network-based augmentation
Liu, C., An, J., Nguyen, X. C., & Balasubramanian, P. (2025) · Energy & Environmental Sustainability · doi:10.1016/j.eesus.2025.100043 · Cited by 15
A Framework for Drift Detection and Adaptation in AI-driven Anomaly and Threat Detection Systems
Lara-Gutierrez, A., Fernández-Gago, C., & Onieva, J. A. (2025) · International Journal of Information Security · doi:10.1007/s10207-025-01118-9 · Cited by 15
“…sampling approaches, such as random undersampling of the majority (benign) class, we observed that under-sampling discards valuable benign samples, affecting model generalization, and often degrades classifier performance by increasing false positives and lowering macro F1, as also stated in [6].”
Analyzing Differentiable Fuzzy Logic Operators
van Krieken, E., Acar, E., & van Harmelen, F. (2021) · Artificial Intelligence · doi:10.1016/j.artint.2021.103602 · Cited by 26
“A second challenge is related to the class imbalance problem (Japkowicz and Stephen, 2002; Buda et al., 2018).”
Background
The Impact of Partial Balance of Imbalanced Dataset on Classification Performance
Li, Q., Zhao, C., He, X., Chen, K., & Wang, R. (2022) · Electronics · doi:10.3390/electronics11091322 · Cited by 15
“Buda et al. [5] proposed two indicators to describe data imbalance and set imbalanced image datasets to verify the impact of imbalanced datasets on CNN under different parameter settings.”
Background
Disease Recognition of Apple Leaf Using Lightweight Multi-Scale Network with ECANet
Yu, H., Cheng, X., Li, Z., Cai, Q., & Bi, C. (2022) · Computer Modeling in Engineering & Sciences · doi:10.32604/cmes.2022.020263 · Cited by 15
Prediction of Smoking Habits From Class-Imbalanced Saliva Microbiome Data Using Data Augmentation and Machine Learning
López, C. D., González, D. M., Vidaki, A., & Kayser, M. (2022) · Frontiers in Microbiology · doi:10.3389/fmicb.2022.886201 · Cited by 16
“This class imbalance leads to spurious high classification accuracy favoring the majority class, while research often focuses on the minority class, and limits the classifier’s generalization (Japkowicz and Stephen, 2002; Abd Elrahman andAbraham, 2013; Ali et al., 2013; Thabtah et al., 2020).”
BackgroundMethod
Deep learning-based automatic-bone-destruction-evaluation system using contextual information from other joints
Miyama, K., Bise, R., Ikemura, S., Kai, K., Kanahori, M., Arisumi, S., Uchida, T., Nakashima, Y., & Uchida, S. (2022) · Arthritis Research & Therapy · doi:10.1186/s13075-022-02914-7 · Cited by 17
“AI: Artificial intelligence; CMC: Carpometacarpal; CMC-M: Carpometacarpal joint of the thumb and multangular; DNN: Deep neural networks; GT: Groundtruth; IP: Interphalangeal; JSN: Joint space narrowing; MCP: Metacarpophalangeal; MIMO: Multiple-input multiple-output; PIP: Proximal interphalangeal; RA: Rheumatoid arthri…”
BackgroundInfluential
High-resolution 3D mapping of cold-water coral reefs using machine learning
de Oliveira, L. M. C., Lim, A., Conti, L. A., & Wheeler, A. J. (2022) · Frontiers in Environmental Science · doi:10.3389/fenvs.2022.1044706 · Cited by 15
Prediction of Highly Imbalanced Semiconductor Chip-Level Defects in Module Tests Using Multimodal Fusion and Logit Adjustment
Cho, H. S., Koo, W., & Kim, H. (2023) · IEEE Transactions on Semiconductor Manufacturing · doi:10.1109/tsm.2023.3283101 · Cited by 16
“One approach is re-sampling, including under-sampling the dominant classes [43], [44] and over-sampling rare classes [45], [46], [47].”
Background
Automatic body part identification in real‐world clinical dermatological images using machine learning
Sitaru, S., Oueslati, T., Schielein, M. C., Weis, J., Kaczmarczyk, R., Rueckert, D., Biedermann, T., & Zink, A. (2023) · JDDG Journal der Deutschen Dermatologischen Gesellschaft · doi:10.1111/ddg.15113 · Cited by 15
“26 The imbalance was corrected using data augmentation as described in the methodology.”
Method
Deep-learning model for predicting physical fitness in possible sarcopenia: analysis of the Korean physical fitness award from 2010 to 2023
Bae, J. H., Seo, J.-W., & Kim, D. (2023) · Frontiers in Public Health · doi:10.3389/fpubh.2023.1241388 · Cited by 15
“15% in our study because of early stopping and using the model checkpoint method, which improved model performance and efficiency (42, 43).”
Method
Subclassified Loss: Rethinking Data Imbalance From Subclass Perspective for Semantic Segmentation
Qiu, S., Cheng, X., Lu, H., Zhang, H., Wan, R., Xue, X., & Pu, J. (2023) · IEEE Transactions on Intelligent Vehicles · doi:10.1109/tiv.2023.3325343 · Cited by 16
“One of the most commonly used methods to address data imbalance is re-sampling [16], [17], [18].”
Method
Addressing class imbalance in deep learning for acoustic target classification
Pala, A., Oleynik, A., Utseth, I., & Handegard, N. O. (2023) · ICES Journal of Marine Science · doi:10.1093/icesjms/fsad165 · Cited by 20
Handling Imbalanced Data for Improved Classification Performance: Methods and Challenges
Abokadr, S., Azman, A., Hamdan, H., & Nasharuddin, N. A. (2023) · International Conference on Emerging Smart Technologies and Applications (eSmarTA) · doi:10.1109/esmarta59349.2023.10293442 · Cited by 15
Euclidpreparation
Leuzzi, L., Meneghetti, M., Angora, G., Metcalf, R. B., Moscardini, L., Rosati, P., Bergamini, P., Calura, F., Clément, B., Gavazzi, R., et al. (2023) · Astronomy and Astrophysics · doi:10.1051/0004-6361/202347244 · Cited by 17
ECOGEN : Bird sounds generation using deep learning
Guei, A., Christin, S., Lecomte, N., & Hervet, É. (2023) · Methods in Ecology and Evolution · doi:10.1111/2041-210x.14239 · Cited by 15
“This leads to class imbalance, a situation that can impair the performance of deep learning models (Buda et al., 2018 ; Johnson & Khoshgoftaar, 2019 ).”
Background
Detecting accounting fraud in family firms: Evidence from machine learning approaches
Rahman, M. J., & Zhu, H. (2023) · Advances in Accounting · doi:10.1016/j.adiac.2023.100722 · Cited by 20
Center-Wise Feature Consistency Learning for Long-Tailed Remote Sensing Object Recognition
Zhao, W., Zhang, Z., Liu, J., Liu, Y., He, Y., & Lu, H. (2024) · IEEE Transactions on Geoscience and Remote Sensing · doi:10.1109/tgrs.2024.3390764 · Cited by 16
“Existing methods solve the long-tailed object recognition through either resampling [15], [16], [17], [18], [19], [43], [44], [45], [46], [47] or cost-sensitive reweighting [13], [20], [21], [22], [23], [24], [25], [48], [49], [50].”
Method
China's first sub-meter building footprints derived by deep learning
Huang, X., Zhang, Z., & Li, J. K. (2024) · Remote Sensing of Environment · doi:10.1016/j.rse.2024.114274 · Cited by 17
Customised-sampling approach for pipe failure prediction in water distribution networks
Latifi, M., Zali, R. B., Javadi, A. A., & Farmani, R. (2024) · Scientific Reports · doi:10.1038/s41598-024-69109-9 · Cited by 15
Unveiling Alzheimer’s Disease Early: A Comprehensive Review of Machine Learning and Imaging Techniques
Hechkel, W., & Helali, A. (2024) · Archives of Computational Methods in Engineering · doi:10.1007/s11831-024-10179-3 · Cited by 15
“Resampling is a common approach that balances class sizes through oversampling or undersam-pling [61].”
AnoOnly: Semi-supervised anomaly detection with the only loss on anomalies
Zhou, Y., Yang, P., Qu, Y., Xu, X., Sun, Z., & Cichocki, A. S. (2024) · Expert Systems with Applications · doi:10.1016/j.eswa.2024.125597 · Cited by 15
“To address this challenge, some recent studies [37,58,38] have attempted to augment supervision volume on anomalies by oversampling strategies [5,14].”
Background
Stress management with HRV following AI, semantic ontology, genetic algorithm and tree explainer
Chatterjee, A., Riegler, M. A., Ganesh, K., & Halvorsen, P. (2025) · Scientific Reports · doi:10.1038/s41598-025-87510-w · Cited by 15
A Hierarchical Approach to Skin Lesion Classification
Kulhalli, R., Savadikar, C., & Garware, B. (2019) · Joint International Conference on Data Science & Management of Data (CODS-COMAD) · doi:10.1145/3297001.3297033 · Cited by 15
Stateful Premise Selection by Recurrent Neural Networks
Piotrowski, B., & Urban, J. (2020) · EPiC series in computing · doi:10.29007/j5hd · Cited by 21
“Oversampling is a general method that often improves the performance of neural architectures on imbalanced data [3].”
Method
Rescaled-LSTM for Predicting Aircraft Component Replacement Under Imbalanced Dataset Constraint
Dangut, M. D., Skaf, Z., & Jennions, I. K. (2020) · Advances in Science and Engineering Technology International Conferences (ASET) · doi:10.1109/aset48392.2020.9118253 · Cited by 15
“Other works that have shown how imbalanced classification can be handled in training neural networks are (Buda, Maki, and Mazurowski 2018; Khan et al. 2018; Lee, Park, and Kim 2016; Lin, Tang, and Yao 2013; Tallón-ballesteros and Hutchison 2017).”
Background
Detection of adverse events leading to inadvertent injury during laparoscopic cholecystectomy using convolutional neural networks
Beyersdorffer, P., Kunert, W., Jansen, K. T., Miller, J., Wilhelm, P., Burgert, O., Kirschniak, A., & Rolinger, J. (2021) · Biomedizinische Technik/Biomedical Engineering · doi:10.1515/bmt-2020-0106 · Cited by 15
Compensating class imbalance for acoustic chimpanzee detection with convolutional recurrent neural networks
Anders, F., Kalan, A. K., Kühl, H. S., & Fuchs, M. (2021) · Ecological Informatics · doi:10.1016/j.ecoinf.2021.101423 · Cited by 15
“We give the following possible explanations for this discrepancy: (1) Studies reporting positive effects of resampling commonly worked with imbalanced training sets, but balanced test sets [40, 41].”
Result
Evaluating the Performance of Data Level Methods Using KEEL Tool to Address Class Imbalance Problem
Upadhyay, K., Kaur, P., & Verma, D. K. (2021) · Arabian Journal for Science and Engineering · doi:10.1007/s13369-021-06377-x · Cited by 15
“To check the performance of these methods, authors considered different distribution of data-sets with various levels of imbalance [77, 80].”
Background
A Method for Class-Imbalance Learning in Android Malware Detection
Guan, J. T., Xu, J., & Mao, B. (2021) · Electronics · doi:10.3390/electronics10243124 · Cited by 15
“For an unbalanced data set, the result will be biased towards the majority class [11,12].”
Background
Efficiency of oversampling methods for enhancing software defect prediction by using imbalanced data
Benala, T. R., & Tantati, K. (2022) · Innovations in Systems and Software Engineering · doi:10.1007/s11334-022-00457-3 · Cited by 15
“Buda et al. [28] reported that in many cases, simple oversampling sufﬁces, that is, many times, one example can be selected randomly, and then, it can be introduced several times into the training set.”
Background
Pattern Recognition Method for Detecting Engineering Errors on Technical Drawings
Джусупова, Р., Banotra, R., Bosch, J., & Olsson, H. H. (2022) · IEEE World AI IoT Congress (AIIoT) · doi:10.1109/aiiot54504.2022.9817294 · Cited by 15
Mutual Learning for Long-Tailed Recognition
Park, C., Yim, J., & Jun, E. (2023) · IEEE Winter Conference on Applications of Computer Vision · doi:10.1109/wacv56688.2023.00270 · Cited by 16
“Under the imbalanced circum-stance, a naively learned neural network is easily dominated by head classes and performs disastrously in tail classes [1].”
BackgroundMethod
Strategies for Tackling the Class Imbalance Problem in Marine Image Classification
Langenkämper, D., van Kevelaer, R., & Nattkemper, T. W. (2018) · Lecture notes in computer science · doi:10.1007/978-3-030-05792-3_3 · Cited by 17
Enhancing Robustness of Deep Neural Networks Against Adversarial Malware Samples: Principles, Framework, and AICS'2019 ChallengePreprint
Li, D., Li, Q., Ye, Y., & Xu, S. (2018) · arXiv (Cornell University) · doi:10.48550/arxiv.1812.08108 · Cited by 18
Elastic-InfoGAN: Unsupervised Disentangled Representation Learning in Class-Imbalanced DataPreprint
Ojha, U., Singh, K. K., Hsieh, C., & Lee, Y. (2019) · arXiv (Cornell University) · doi:10.48550/arxiv.1910.01112 · Cited by 16
“To alleviate this issue, researchers have proposed re-sampling (Chawla et al. (2002); He et al. (2008); Shen et al. (2016); Buda et al. (2018); Zou et al. (2018)) and class reweighting techniques (Ting (2000); Huang et al. (2016); Dong et al. (2017); Mahajan et al. (2018)) to oversample rare…”
Method
Transfer Learning for Oral Cancer Detection using Microscopic ImagesPreprint
Palaskar, R. J., Vyas, R., Khedekar, V. B., Palaskar, S. J., & Sahu, P. (2020) · arXiv (Cornell University) · doi:10.48550/arxiv.2011.11610 · Cited by 15
“Handling Class Imbalance We follow three common techniques of handling class imbalance (Buda et al., 2018): (1) Oversampling where images from the minority class are randomly sampled until the cardinality of both classes is same, and equal to the majority class.”
Method
Advances in liver, liver lesion, hepatic vasculature, and biliary segmentation: a comprehensive review of traditional and deep learning approaches
Sethia, K., Strakoš, P., Jaros, M., Kubíček, J., Roman, J., Penhaker, M., & Řı́ha, L. (2025) · Artificial Intelligence Review · doi:10.1007/s10462-025-11310-x · Cited by 19
Tap water fingerprinting using a convolutional neural network built from images of the coffee-ring effect
Li, X., Sanderson, A. R., Allen, S. S., & Lahr, R. H. (2019) · The Analyst · doi:10.1039/c9an01624d · Cited by 14
Balancing Data on Deep Learning-Based Proteochemometric Activity Classification
Rio, A. L.-D., Picart‐Armada, S., & Perera-Lluna, A. (2021) · Journal of Chemical Information and Modeling · doi:10.1021/acs.jcim.1c00086 · Cited by 15
Deep learning applied to breast imaging classification and segmentation with human expert intervention
Wilding, R., Sheraton, V. M., Soto, L., Chotai, N. C., & Tan, E. Y. (2022) · Journal of Ultrasound · doi:10.1007/s40477-021-00642-3 · Cited by 13
“Due to class imbalance within the dataset, we used oversampling as presented elsewhere as a method to handle class imbalance [24].”
Method
A comparative study on class-imbalanced gas turbine fault diagnosis
Bai, M., Liu, J., Long, Z., Luo, J., & Yu, D. (2022) · Proceedings of the Institution of Mechanical Engineers Part G Journal of Aerospace Engineering · doi:10.1177/09544100221107252 · Cited by 14
“The class-imbalance problem is caused by the lack of fault samples and can greatly reduce the fault diagnosis accuracy of fault samples.(41) The class-imbalanced problem is quite important but often neglected in gas turbine fault diagnosis.”
Background
Teaching Teachers First and Then Student: Hierarchical Distillation to Improve Long-Tailed Object Recognition in Aerial Images
Zhao, W., Liu, J., Liu, Y., Zhao, F., He, Y., & Lu, H. (2022) · IEEE Transactions on Geoscience and Remote Sensing · doi:10.1109/tgrs.2022.3177853 · Cited by 14
Interpatient Heartbeat Classification Using Modified Residual Attention Network With Two-Phase Training and Assistant Decision
Wang, Y., Zhou, G., & Yang, C. (2022) · IEEE Transactions on Instrumentation and Measurement · doi:10.1109/tim.2022.3232646 · Cited by 18
“…network [32], [33], [34], [35], guided evolutionary synthesizer [36], focal loss (FL) [37], and batch-weighted loss [9]) have focused on data balance, summarized as resampling (data level) methods and algorithmic-level methods, the details of which can be found in the systematic study of [38].”
BackgroundMethod
Transference of Training for a DNN to Complete the Aerial Refueling Task
Costello, D. H., Mwaffo, V., & Miller, D. (2023) · AIAA SciTech Forum · doi:10.2514/6.2023-0192 · Cited by 14
An Optimal Transport View of Class-Imbalanced Visual Recognition
Jin, L., Lang, D., & Lei, N. (2023) · International Journal of Computer Vision · doi:10.1007/s11263-023-01831-9 · Cited by 14
“One of the most widely used class re-balancing methods is data re-sampling, mainly over-sample tail class images (Han et al., 2005; Chawla et al., 2002) or under-sample head class images (Liu et al., 2009; Buda et al., 2018).”
Method
Adaptive undersampling and short clip-based two-stream CNN-LSTM model for surgical phase recognition on cholecystectomy videos
Lee, S., Kim, G., Hwang, Y. N., Kwon, J. Y., & Kim, S. (2023) · Biomedical Signal Processing and Control · doi:10.1016/j.bspc.2023.105637 · Cited by 16
Physical Insights From the Multidecadal Prediction of North Atlantic Sea Surface Temperature Variability Using Explainable Neural Networks
Liu, G., Wang, P., & Kwon, Y. (2023) · Geophysical Research Letters · doi:10.1029/2023gl106278 · Cited by 14
Developing a web application for Chinese calligraphy learners using convolutional neural network and scale invariant feature transform
Wong, A. K. L., So, J. C. H., & Ng, Z. T. B. (2024) · Computers and Education Artificial Intelligence · doi:10.1016/j.caeai.2024.100200 · Cited by 14
Virtual Classification: Modulating Domain-Specific Knowledge for Multidomain Crowd Counting
Guo, M., Chen, B., Yan, Z., Wang, Y., & Ye, Q. (2024) · IEEE Transactions on Neural Networks and Learning Systems · doi:10.1109/tnnls.2024.3350363 · Cited by 15
“As shown in Table VI, the oversampling strategy suffers from heavy overfitting to tail classes especially on small datasets, e.g., SHA and SHB, and performance decrease on QNRF and NWPU, which is consistent with the result in [84].”
Enhancing Multi-species Liver Microsomal Stability Prediction through Artificial Intelligence
Long, T.-Z., Jiang, D., Shi, S., Deng, Y., Wang, W., & Cao, D. (2024) · Journal of Chemical Information and Modeling · doi:10.1021/acs.jcim.4c00159 · Cited by 14
Smart contract vulnerability detection using wide and deep neural network
Osei, S. B., Ma, Z., & Huang, R. (2024) · Science of Computer Programming · doi:10.1016/j.scico.2024.103172 · Cited by 14
A prototype-assisted clustered federated learning for big data security and privacy preservation
Jiang, Y., Wang, D., Song, B., & Du, X. (2024) · Future Generation Computer Systems · doi:10.1016/j.future.2024.07.032 · Cited by 16
Data Imbalance, Uncertainty Quantification, and Transfer Learning in Data‐Driven Parameterizations: Lessons From the Emulation of Gravity Wave Momentum Transport in WACCM
Sun, Y. Q., Pahlavan, H. A., Chattopadhyay, A., Hassanzadeh, P., Lubis, S. W., Alexander, M. J., Gerber, E. P., Sheshadri, A., & Guan, Y. (2024) · Journal of Advances in Modeling Earth Systems · doi:10.1029/2023ms004145 · Cited by 19
“Data imbalance is a well‐known problem in the ML literature, especially in the context of classification tasks (e.g., Ando & Huang, 2017; Buda et al., 2018; Chawla et al., 2004; Huang et al., 2016; Japkowicz & Stephen, 2002; Johnson & Khoshgoftaar, 2019; G. Wu & Chang, 2003).”
Time-guided convolutional neural networks for spatiotemporal urban flood modelling
Wang, Z., Lyu, H., Fu, G., & Zhang, C. (2024) · Journal of Hydrology · doi:10.1016/j.jhydrol.2024.132250 · Cited by 15
Self‐Attention Mechanisms‐Based Laryngoscopy Image Classification Technique for Laryngeal Cancer Detection
Kang, Y., Yang, L., Hu, Y., Xu, K. L., Cai, L., Hu, B., & Lu, X. (2024) · Head & Neck · doi:10.1002/hed.27999 · Cited by 15
“Moreover, due to the class imbalance in our dataset, although we employed several techniques to address this issue, the AUC appeared to be a more reliable metric compared to the F 1 score or accuracy, as it is less affected by data imbalance [26].”
Reparameterization convolutional neural networks for handling imbalanced datasets in solar panel fault classification
Guo, J., Chong, C. F., Abreu, P. H., Mao, C., Li, J., Lam, C., & Ng, B. K. (2025) · Engineering Applications of Artificial Intelligence · doi:10.1016/j.engappai.2025.110541 · Cited by 14
Vision-Based Crack Detection using Transfer Learning in Metal Forming Processes
Mittel, D., & Kerber, F. (2019) · IEEE International Conference on Emerging Technologies and Factory Automation (ETFA) · doi:10.1109/etfa.2019.8869084 · Cited by 14
“Referring to [19] oversampling is the best choice in most cases to deal with imbalanced datasets for performance and for generalization.”
Background
Multiscale Residual Convolution Neural Network and Sector Descriptor-Based Road Detection Method
Dai, J.-G., Du, Y., Zhu, T., Wang, Y., & Gao, L. (2019) · IEEE Access · doi:10.1109/access.2019.2956725 · Cited by 14
DeepSSV: detecting somatic small variants in paired tumor and normal sequencing data with convolutional neural network
Meng, J., Victor, B., He, Z., Liu, H., & Jiang, T. (2020) · Briefings in Bioinformatics · doi:10.1093/bib/bbaa272 · Cited by 14
JigsawNet: Shredded Image Reassembly Using Convolutional Neural Network and Loop-Based Composition
Le, C., & Li, X. S. (2019) · Civil War Book Review · Cited by 14
A Hybrid VDV Model for Automatic Diagnosis of Pneumothorax Using Class-Imbalanced Chest X-Rays Dataset
Iqbal, T., Shaukat, A., Akram, M. U., Muzaffar, A. W., Mustansar, Z., & Byun, Y.-C. (2022) · IEEE Access · doi:10.1109/access.2022.3157316 · Cited by 16
“Buda et al [21] compared the performance of CNNs using multiple approaches including oversampling, under-sampling, and”
BackgroundInfluential
Dealing With Highly Unbalanced Sidescan Sonar Image Datasets for Deep Learning Classification Tasks
Steiniger, Y., Stoppe, J., Meisen, T., & Kraus, D. (2020) · Global Oceans 2020: Singapore – U.S. Gulf Coast · doi:10.1109/ieeeconf38699.2020.9389373 · Cited by 14
“The balance between the number of samples of classes in a dataset has shown to be an important property considering classiﬁcation with deep learning models [7].”
Background
Label-free imaging and classification of live P. falciparum enables high performance parasitemia quantification without fixation or staining
Lebel, P. M., Dial, R. L., Vemuri, V. N. P., Garcia, V. E., DeRisi, J. L., & Gómez-Sjöberg, R. (2021) · PLoS Computational Biology · doi:10.1371/journal.pcbi.1009257 · Cited by 14
“order to re-normalize the training weights to account for class imbalance [38].”
Background
Multi-Scale Conditional Generative Adversarial Network for Small-Sized Lung Nodules Using Class Activation Region Influence Maximization
Ann, K., Jang, Y., Shim, H., & Chang, H. (2021) · IEEE Access · doi:10.1109/access.2021.3116034 · Cited by 14
Improving Rare-Class Recognition of Marine Plankton with Hard Negative Mining
Walker, J. L., & Orenstein, E. C. (2021) · IEEE International Conference on Computer Vision Workshops · doi:10.1109/iccvw54120.2021.00410 · Cited by 14
“The combined abundance of objects from the target classes is often much smaller than the prevalence of all other objects that form the background, the so-called class imbalance problem [27, 7, 6, 17].”
BackgroundMethod
Rectifying the Data Bias in Knowledge Distillation
Liu, B., Zhang, S., Song, G., You, H., & Liu, Y. (2021) · IEEE International Conference on Computer Vision Workshops · doi:10.1109/iccvw54120.2021.00171 · Cited by 14
Jumping into the artistic deep end: building the catalogue raisonné
Dobbs, T., Benedict, A., & Raś, Z. W. (2022) · AI & Society · doi:10.1007/s00146-021-01370-2 · Cited by 15
“From a measurement perspective, research shows that micro-bal-anced accuracy based on true-positive rate (TPR) and false-positive rate (FPR) is a good predictor when there is a concern for under-represented classes (Grandini et al. 2020).”
Background
Identifying patients with atrial fibrillation during sinus rhythm on ECG: Significance of the labeling in the artificial intelligence algorithm
Suzuki, S., Motogi, J., Nakai, H., Matsuzawa, W., Takayanagi, T., Umemoto, T., Hirota, N., Hyodo, A., Satoh, K., Otsuka, T., et al. (2022) · IJC Heart & Vasculature · doi:10.1016/j.ijcha.2022.100954 · Cited by 16
“The reason for sampling with a 5-second duration was because we employed oversampling to balance the number of samples between AF and SR labels [9,10].”
Method
AFE-MERT: imbalanced text classification with abstract feature extraction
Okkalioglu, M., & Okkalioglu, B. D. (2022) · Applied intelligence (Boston) · doi:10.1007/s10489-021-02983-2 · Cited by 10
“Formally, the imbalance ratio, IR, is the ratio between the class with the maximum number of samples and the class with the minimum number of samples [4, 22] as given in (5) where | C i | represents the number of samples that class c i has.”
Background
A 1D-SP-Net to Determine Early Drought Stress Status of Tomato (Solanum lycopersicum) with Imbalanced Vis/NIR Spectroscopy Data
Tu, Y.-K., Kuo, C., Fang, S.-L., Chen, H., Chi, M.-K., Yao, M.-H., & Kuo, B. (2022) · Agriculture · doi:10.3390/agriculture12020259 · Cited by 16
Machine Vision Assisted Identification of the Lung Adenocarcinoma Category and High-Risk Tumor Area Based on CT Images
Chen, L., Qi, H., Lu, D., Zhai, J., Cai, K., Wang, L., Liang, G., & Zhang, Z. (2021) · Patterns · doi:10.1016/j.patter.2022.100464 · Cited by 15
“Traditional imbalanced learning techniques, 12–14 such as the synthetic minority oversampling technique (SMOTE), 15 also do not work ideally, because lung patterns should remain clear and analyzable clinically to explain the relationship between CT imaging features and results of pathological…”
Method
Smote vs. Random Undersampling for Imbalanced Data - Car Ownership Demand Model
Chaipanha, W., & Kaewwichian, P. (2022) · Communications - Scientific letters of the University of Zilina · doi:10.26552/com.c.2022.3.d105-d115 · Cited by 14
“However, the TPR percentage of each area-type was still low (the lowest = 44.3 % for the total area type), in which the data analysis indicated that this problem had occurred due to the imbalanced data as the Imbalanced Ratio (IR) [28].”
BackgroundInfluential
Locoregional recurrence prediction in head and neck cancer based on multi-modality and multi-view feature expansion
Wang, R., Guo, J., Zhou, Z., Wang, K., Gou, S., Xu, R., Sher, D. J., & Wang, J. (2022) · Physics in Medicine and Biology · doi:10.1088/1361-6560/ac72f0 · Cited by 14
“In an epoch, through random selection ofmultiple batches, positive samples are oversampled (Buda et al 2018, Byrd and Lipton 2019) tomake up theminority class samples.”
Background
SaR: Self-adaptive Refinement on Pseudo Labels for Multiclass-Imbalanced Semi-supervised Learning
Lai, Z., Wang, C., Cheung, S.-C. S., & Chuah, C. (2022) · IEEE/CVF Conference on Computer Vision and Pattern Recognition Workshops (CVPRW) · doi:10.1109/cvprw56347.2022.00454 · Cited by 15
“Recent approaches include re-weighting the loss function [2,9,10,14,21,39] and re-sampling the training data [7, 11, 19, 20, 33, 48].”
BackgroundMethod
Ocular diseases classification using a lightweight CNN and class weight balancing on OCT images
Gour, N., & Khanna, P. (2022) · Multimedia Tools and Applications · doi:10.1007/s11042-022-13617-1 · Cited by 14
“If the number of samples in one class is significantly higher than in the other classes, then the dataset is termed imbalanced [3].”
Background
Improving Smiling Detection with Race and Gender Diversity
Ryu, H. J., Mitchell, M., & Adam, H. (2017) · Cited by 14
Detection of Algorithmically Generated Domain Names Using the Recurrent Convolutional Neural Network with Spatial Pyramid Pooling
Liu, Z., Zhang, Y., Chen, Y., Fan, X., & Dong, C. (2020) · Entropy · doi:10.3390/e22091058 · Cited by 14
“Considering that the neural model requires a large-scale dataset for parameter tuning and is sensitive to sample distribution imbalance [27], the sample equalization module employs the borderline synthetic minority over-sampling algorithm (Borderline-SMOTE) to optimize the sample balance in the dataset.”
Method
Machine Learning of Spatial Data
Nikparvar, B., & Thill, J. (2021) · ISPRS International Journal of Geo-Information · doi:10.3390/ijgi10090600 · Cited by 163
“This problem is known as intra-class imbalance [87].”
Background
A Scoping Review of Arabic Natural Language Processing for Mental Health
Alasmari, A. (2025) · Healthcare · doi:10.3390/healthcare13090963 · Cited by 15
“Class weighting adjusts the loss function to penalize misclassifications of rare conditions [57], while data augmentation techniques like back-translation and synthetic data generation increase underrepresented samples [58].”
Random forest is the best species predictor for a community of insectivorous bats inhabiting a mountain ecosystem of central Mexico
Ayala‐Berdón, J., Medina-Bello, K. I., López‐Cuamatzi, I. L., Vázquez-Fuerte, R., G., M. C. M., Orozco-Lugo, L., Íñiguez‐Dávalos, L. I., Guillén-Servent, A., & Martı́nez–Gómez, M. (2020) · Bioacoustics · doi:10.1080/09524622.2020.1835539 · Cited by 13
“In this condition, balancing data (either by weighting or resampling by over-sampling or under-sampling the data) is recommended (Buda et al. 2018).”
Background
Approximating Three‐Dimensional (3‐D) Transport of Atmospheric Pollutants via Deep Learning
Xu, J. Z., Zhang, H. R., Cheng, Z., Liu, J. Y., Xu, Y., & Wang, Y. C. (2022) · Earth and Space Science · doi:10.1029/2022ea002338 · Cited by 13
“…known to be well mixed while the pollutant's concentration above the PBL can be much lower which cause imbalance in the overall data (Kim et al., 2015; Lu et al., 2019), that will tremendously challenge the approximating ability of the deep learning surrogate (Buda et al., 2018; Liu et al., 2019).”
Background
Efficient Arithmetic Block Identification With Graph Learning and Network-Flow
Wang, Z., He, Z., Bai, C., Yang, H., & Yu, B. (2022) · IEEE Transactions on Computer-Aided Design of Integrated Circuits and Systems · doi:10.1109/tcad.2022.3227815 · Cited by 13
“We adopt the basic version of it, called random minority oversampling, which supplements the training data with multiple copies of some of the minority classes [37].”
BackgroundMethod
Machine Learning for Prediction of Wait Times in Outpatient Clinic
Joseph, J., Senith, S., Kirubaraj, A. A., & Ramson, S. R. J. (2022) · Procedia Computer Science · doi:10.1016/j.procs.2022.12.026 · Cited by 13
Data augmentation and deep neural network classification based on ship radiated noise
Xie, Z., Lin, R., Wang, L., Zhang, A., Lin, J., & Tang, X. (2023) · Frontiers in Marine Science · doi:10.3389/fmars.2023.1113224 · Cited by 13
“The multi-scale analysis algorithm is developed by Costa, and can be used to estimate the complexity of the original data at different scales Chen et al. (2019). Based on the concept of multi-scale analysis, multi-scale permutation entropy (MPE) is proposed by Aziz and Arif (2005).”
Method
A novel method to predict slip resistance of winter footwear using a convolutional neural network
Lau, K., Fernie, G. R., & Fekr, A. R. (2023) · Footwear Science · doi:10.1080/19424280.2023.2198987 · Cited by 13
“These methods can be divided into three categories: (1) data level methods that change the training data, (2) classifier level methods that adjust the training algorithms, and (3) combinations of methods (1) and (2) (Buda et al., 2018; Haixiang et al., 2017; H.”
Method
Hybrid Optimization based Feature Selection with DenseNet Model for Heart Disease Prediction
Krishnan, V. G., Saradhi, M. V. V., Kumar, S. S., Dhanalakshmi, G., Pushpa, P., & Vijayaraja, V. (2023) · International Journal of Electrical and Electronics Research · doi:10.37391/ijeer.110203 · Cited by 13
“As a result of the imbalanced data, classification mistakes occur when ML models are being trained [29].”
Background
Context-aggregator: An approach of loss- and class imbalance-aware aggregation in federated learning
Abbas, Q., Malik, K. M., Saudagar, A. K. J., & Khan, M. B. (2023) · Computers in Biology and Medicine · doi:10.1016/j.compbiomed.2023.107167 · Cited by 15
Investigation of machine learning algorithms for taxonomic classification of marine metagenomes
Park, H., Lim, S. J., Cosme, J., O’Connell, K. A., Sandeep, J., Gayanilo, F. C., Cutter, G. R., Montes, E., Nitikitpaiboon, C., Fisher, S., et al. (2023) · Microbiology Spectrum · doi:10.1128/spectrum.05237-22 · Cited by 13
“The model then has high recognition of overrepresented classes but mispredicts underrepresented classes during testing (37).”
Background
Image Anomaly Detection with Capsule Networks and Imbalanced Datasets
Piciarelli, C., Mishra, P., & Foresti, G. L. (2019) · Lecture Notes in Computer Science · doi:10.1007/978-3-030-30642-7_23 · Cited by 10
“detection with capsule networks and imbalanced datasets 3 of new data. In both cases the idea is to use a pre-processing step to make the dataset balanced before applying any classification algorithm [4]. Regardingsemi-supervised orunsupervisedapproaches,earlyworksadopted techniques such as Deep Belief Networks [20] f…”
Method
A Comprehensive Study of Class Incremental Learning Algorithms for Visual Tasks
Belouadah, E., Popescu, A., & Kanellos, I. (2020) · arXiv · doi:10.48550/arxiv.2011.01844 · Cited by 235
“The well-known thresholding method [8] is used to reduce the classiﬁcation bias between past and new classes.”
MethodInfluential

Authors: Buda, M., Maki, A., & Mazurowski, M. A.
Affiliations: Duke University (US)
KTH Royal Institute of Technology (SE)
Corresponding author: Buda, Mateusz
ORCID: Buda, Mateusz
Maki, Atsuto
Mazurowski, Maciej A.
Venue: Neural Networks
Publisher: Elsevier BV
Published: 2018-10
Volume: 106
Pages: 249-259
ISSN: 0893-6080 (ISSN-L)
DOI: 10.1016/j.neunet.2018.07.011
OpenAlex: W2767106145
Cited by: 3,088
References: 83
Citations by year: 2026: 459 · 2025: 507 · 2024: 371 · 2023: 405 · 2022: 405 · 2021: 380 · 2020: 308 · 2019: 189 · 2018: 58
FWCI: 131.58
Type: journal article
Language: en
Open access: green
arxiv.org
Licence: www.elsevier.com/tdm/userlicense/1.0/ (tdm)
www.elsevier.com/legal/tdmrep-license (tdm)
Locations: Neural Networks (publishedVersion)
PubMed (publishedVersion)
arXiv (Cornell University) (submittedVersion)
KTH Publication Database DiVA (KTH Royal Institute of Technology) (submittedVersion, other-oa)
arXiv (Cornell University)
Topics: Physical Sciences › Computer Science › Artificial Intelligence › Imbalanced Data Classification Techniques
Physical Sciences › Computer Science › Computer Vision and Pattern Recognition › Advanced Neural Network Applications
Physical Sciences › Computer Science › Artificial Intelligence › Machine Learning and Data Classification
Keywords: class imbalance, convolutional neural networks, oversampling, undersampling, two-stage training, thresholding, area under the receiver operating characteristic, multi-class classification, overfitting, MNIST, CIFAR-10, ImageNet
MeSH terms: Machine Learning, Humans, Probability, ROC Curve, Neural Networks, Computer

About these checks

Registry, retraction, Cites and statistics results are computed, not judged, with one exception: a printed reference that no index lists is read once by a model before it counts as a citation error. Claims are extracted by a model and kept only when their quote appears verbatim in the text; each cited source is then checked, with a model’s help, for whether it says what the claim relies on. Those verdicts can be wrong — read the claim’s quote and the cited source before relying on one.

Statistics and arithmetic are recomputed, not judged.

Confidence bands are shown only for fields calibrated on our annotated test set. None is calibrated yet, so no band is shown.

What counts as a claim ›How we measure confidence ›How we check this ›

Are you an author, or do you see a problem with this page? Request a review

A systematic study of the class imbalance problem in convolutional neural networks

Abstract

Overview

Integrity signals

Verification

Claims the paper makes

Claims other papers make about this one

Citations

Cites

Cited by

Related

Metadata

Record

About these checks