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 accessAbstract
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.
Overview
Integrity signals
Registry
Cites
Statistics
Verification
Claims the paper makes
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?
In the citing paper
“A memory of the past is needed to rectify past classes' scores.”
From A Comprehensive Study of Class Incremental Learning Algorithms for Visual Tasks (2020) · cites this paper as [12]
In the citing paper
“Here, we introduced a competitive method where classification bias in favor of new classes is reduced by using prior class probabilities [12].”
From A Comprehensive Study of Class Incremental Learning Algorithms for Visual Tasks (2020) · cites this paper as [12]
In the citing paper
“Finally, the highly-discussed dimension target class balance pays tribute to the technical requirements of ML140,141,144,150,159.”
From The METRIC-framework for assessing data quality for trustworthy AI in medicine: A systematic review (2024) · cites this paper as [141]
Citations
Cites
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.”
BackgroundVery 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].”
MethodImageNet 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].”
MethodGradient-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].”
MethodImageNet 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].”
MethodSMOTE 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].”
BackgroundMethodDelving 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].”
MethodVisualizing 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].”
BackgroundBatch Normalization: Accelerating Deep Network Training by Reducing Internal Covariate ShiftPreprint
Ioffe, S. (2024) · arXiv (Cornell University) · doi:10.57702/o9raffed · Cited by 15,273
Cited by
Sorted by citations
Items 701–800 of 1,001
Tang, M., Chen, Y., Wu, H., Zhao, Q., Long, W., Sheng, V. S., & Yi, J. (2021) · Frontiers in Energy Research · doi:10.3389/fenrg.2021.686616 · Cited by 23
“In the actual operation of wind turbines, the number of fault samples is much smaller than the number of normal samples, which is characteristic of typical imbalanced classi fi cation problems (Malik and Mishra, 2016; Buda et al., 2018; Longting et al., 2019).”
BackgroundSemi-supervised target classification in multi-frequency echosounder data
Choi, C., Kampffmeyer, M., Handegard, N. O., Salberg, A.-B., Brautaset, O., Eikvil, L., & Jenssen, R. (2021) · ICES Journal of Marine Science · doi:10.1093/icesjms/fsab140 · Cited by 23
Chai, S. S., Goh, K. L., Cheah, W. L., Chang, Y. H. R., & Ng, G. W. (2022) · Applied Sciences · doi:10.3390/app12031600 · Cited by 23
Chen, D., Shen, Y., Zhang, H., & Torr, P. H. S. (2022) · International Joint Conference on Artificial Intelligence · doi:10.24963/ijcai.2022/114 · Cited by 28
Prototype-based classifier learning for long-tailed visual recognition
Wei, X.-S., Xu, S.-L., Chen, H., Xiao, L., & Peng, Y. (2022) · Science China Information Sciences · doi:10.1007/s11432-021-3489-1 · Cited by 22
“Re-sampling methods as one of the most important class re-balancing strategies could be divided into two types: (1) Over-sampling by simply repeating data of minority classes [6, 28, 29] and (2) under-sampling by abandoning data of dominant classes [7,28,30].”
BackgroundME-D2N: Multi-Expert Domain Decompositional Network for Cross-Domain Few-Shot Learning
Fu, Y., Xie, Y. H., Fu, Y., Chen, J., & Jiang, Y. (2022) · ACM International Conference on Multimedia (ACM MM) · doi:10.1145/3503161.3547995 · Cited by 22
“Main-stream methods include: 1) re-sampling based methods [2, 3, 19] which under-sample the head classes or over-sample the tail classes; 2) re-weighting based methods [5, 21, 22, 50] which assign different weights for classes or instances; 3) two-stage fine-tuning based methods [7, 26, 57] which train the representat…”
BackgroundMethodSemantic Segmentation and 3D Reconstruction of Concrete Cracks
Shokri, P., Shahbazi, M., & Nielsen, J. (2022) · Remote Sensing · doi:10.3390/rs14225793 · Cited by 25
Sonar Images Classification While Facing Long-Tail and Few-Shot
Jiao, W., & Zhang, J. (2022) · IEEE Transactions on Geoscience and Remote Sensing · doi:10.1109/tgrs.2022.3211847 · Cited by 23
The power of deep learning for intelligent tumor classification systems: A review
Chandni, Sachdeva, M., & Kushwaha, A. K. S. (2023) · Computers & Electrical Engineering · doi:10.1016/j.compeleceng.2023.108586 · Cited by 24
Deep ensemble learning enables highly accurate classification of stored red blood cell morphology
Routt, A. H., Yang, N., Piety, N. Z., Lu, M., & Shevkoplyas, S. S. (2023) · Scientific Reports · doi:10.1038/s41598-023-30214-w · Cited by 23
Real-Time Pose Tracking for a Continuum Guidewire Robot Under Fluoroscopic Imaging
Ravigopal, S. R., Sarma, A., Brumfiel, T. A., & Desai, J. P. (2023) · IEEE Transactions on Medical Robotics and Bionics · doi:10.1109/tmrb.2023.3260273 · Cited by 24
“The imbalance ratio [33] of a dataset is designated as the maximum number of pixels that comprise the largest classification set (i.”
BackgroundAutomated seismic semantic segmentation using attention U-Net
AlSalmi, H., & Elsheikh, A. H. (2023) · Geophysics · doi:10.1190/geo2023-0149.1 · Cited by 25
NCL++: Nested Collaborative Learning for long-tailed visual recognition
Tan, Z., Li, J., Du, J., Wan, J., Lei, Z., & Guo, G. (2023) · Pattern Recognition · doi:10.1016/j.patcog.2023.110064 · Cited by 28
Baik, J. S., Yoon, I. Y., & Choi, J. W. (2023) · Pattern Recognition · doi:10.1016/j.patcog.2023.110107 · Cited by 23
“Popularly used methods include re-sampling [1, 2, 4, 9, 10, 26] and re-weighting [3,7,13,14,21,25,27,34].”
MethodJensen–Shannon divergence based novel loss functions for Bayesian neural networks
Thiagarajan, P., & Ghosh, S. (2024) · Neurocomputing · doi:10.1016/j.neucom.2024.129115 · Cited by 28
“Background Despite the widespread success of deep neural networks (DNNs) and convolutional neural networks (CNNs) in numerous science and engineering applications [1–10], they suffer from overfitting when the data set is small, noisy or biased [3, 11].”
JellyNet: The convolutional neural network jellyfish bloom detector
Mcilwaine, B., & Casado, M. R. (2021) · International Journal of Applied Earth Observation and Geoinformation · doi:10.1016/j.jag.2020.102279 · Cited by 24
“Image chips were extracted using the digital image cropping tool Greenshot (Braun et al., 2020) in combination with the image viewer IrfanView (Skiljan, 2019) to aid speed of processing the chips from the original raw images.”
MethodCost-Sensitive Convolution based Neural Networks for Imbalanced Time-Series ClassificationPreprint
Geng, Y., & Luo, X. (2018) · arXiv (Cornell University) · doi:10.48550/arxiv.1801.04396 · Cited by 23
PCAfold: Python software to generate, analyze and improve PCA-derived low-dimensional manifolds
Zdybał, K., Armstrong, E., Parente, A., & Sutherland, J. C. (2020) · SoftwareX · doi:10.1016/j.softx.2020.100630 · Cited by 23
“Several authors report an advantage of changing sampling densities of train and test data when using data analysis or machine learning techniques [8,9].”
BackgroundBüttner, M., Schneider, L., Krasowski, A., Pitchika, V., Krois, J., Meyer‐Lueckel, H., & Schwendicke, F. (2024) · Journal of Dentistry · doi:10.1016/j.jdent.2024.105063 · Cited by 22
Khattak, A., Zhang, J., Chan, P. W., & Chen, F. (2024) · Expert Systems with Applications · doi:10.1016/j.eswa.2024.124399 · Cited by 22
Gürcan, F., & Soylu, A. (2024) · Cancers · doi:10.3390/cancers16234046 · Cited by 22
“Studies have shown that traditional oversampling methods like SMOTE and ADASYN, while effective to some degree, may not fully capture the complexity of real data distributions in medical contexts [9,11–15].”
Oliveira-Filho, A., Comeau, M., Cave, J., Nasr, C., Côté, P., & Tahan, A. S. (2024) · Energies · doi:10.3390/en18010059 · Cited by 28
“…of the number of data points in the majority class to those in the minority class (majority class number of points:minority class number of points) or the ratio between the datasets’ duration (duration covered by the majority class dataset:duration covered by the minority class dataset) [24,26].”
FG-SMOTE: Towards Fair Node Classification with Graph Neural Network
Wang, Z., Yin, Z., Zhang, Y., Yang, L., Zhang, T., Pissinou, N., Cai, Y., Hu, S., Li, Y., Zhao, L., et al. (2025) · ACM SIGKDD Explorations Newsletter · doi:10.1145/3715073.3715082 · Cited by 23
“For instance, Oversampling [4] amplifies the representation of minority nodes by directly duplicating their embeddings from real datasets.”
Liu, Y., Qi, H., & Lu, X. (2025) · Computer Assisted Language Learning · doi:10.1080/09588221.2025.2518430 · Cited by 29
Cyber Secure Man-in-the-Middle Attack Intrusion Detection Using Machine Learning Algorithms
Jayapandian, N. (2019) · Advances in computational intelligence and robotics book series · doi:10.4018/978-1-5225-9687-5.ch011 · Cited by 22
Soft trees with neural components as image-processing technique for archeological excavations
Woźniak, M., & Połap, D. (2020) · Personal and Ubiquitous Computing · doi:10.1007/s00779-019-01292-3 · Cited by 22
A Tiny CNN Architecture for Identifying Bat Species from Echolocation Calls
Zualkernan, I. A., Judas, J., Mahbub, T., Bhagwagar, A., & Chand, P. (2020) · IEEE/ITU International Conference on Artificial Intelligence for Good (AI4G) · doi:10.1109/ai4g50087.2020.9311084 · Cited by 22
Kuklyte, J., Fitzgerald, J., Nelissen, S. R., Wei, H., Whelan, A., Power, A., Ahmad, A., Miarka, M., Gregson, M., Maxwell, M., et al. (2021) · Toxicologic Pathology · doi:10.1177/0192623320986423 · Cited by 24
“Buda et al 52 The results of introducing class balancing for lesion detection in 5 different nonclinical tissues are demonstrated in Figure 5.”
BackgroundRealistic preterm prediction based on optimized synthetic sampling of EHG signal
Xu, J., Chen, Z., Zhang, J., Lu, Y., Yang, X. C., & Pumir, A. (2021) · Computers in Biology and Medicine · doi:10.1016/j.compbiomed.2021.104644 · Cited by 24
“In addition to the sample abundance and feature quality already mentioned, the distribution pattern of samples among classes is also an important factor affecting classification performance [12, 13].”
BackgroundQuan, J., Miyake, Y., & Nozawa, T. (2021) · Sensors · doi:10.3390/s21165317 · Cited by 23
Deep learning‐based automated mitosis detection in histopathology images for breast cancer grading
Mathew, T., Ajith, B., Kini, J. R., & Rajan, J. (2022) · International Journal of Imaging Systems and Technology · doi:10.1002/ima.22703 · Cited by 22
“53,54 In this binary classification problem, the positive class of mitotic cells is far less compared to non-mitotic cells and mimics.”
BackgroundFine-grained identification of vehicle loads on bridges based on computer vision
Zhu, J., Li, X., & Zhang, C. (2022) · Journal of Civil Structural Health Monitoring · doi:10.1007/s13349-022-00552-w · Cited by 22
“To obtain a reliable detection model, the data augmentation at data level [49] are used.”
MethodYamashita, R., Kapoor, T., Alam, M. N., Galimzianova, A., Syed, S. A., Akdogan, M. U., Alkım, E., Wentland, A. L., Madhuripan, N., Goff, D. J., et al. (2022) · Radiology Artificial Intelligence · doi:10.1148/ryai.210174 · Cited by 22
Iterative Annotation of Biomedical NER Corpora with Deep Neural Networks and Knowledge Bases
Silvestri, S., Gargiulo, F., & Ciampi, M. (2022) · Applied Sciences · doi:10.3390/app12125775 · Cited by 24
“An imbalanced class in the training set could limit the performance of a DNN trained with such corpora [50].”
BackgroundCalibrated Focal Loss for Semantic Labeling of High-Resolution Remote Sensing Images
Bai, H., Cheng, J., Su, Y., Liu, S., & Liu, X. (2022) · IEEE Journal of Selected Topics in Applied Earth Observations and Remote Sensing · doi:10.1109/jstars.2022.3197937 · Cited by 22
Honda, S., Yano, K., Tanaka, E., Ikari, K., & Harigai, M. (2022) · Lara D. Veeken · doi:10.1093/rheumatology/keac586 · Cited by 22
“These imbalanced datasets result in poor generalized performance [5].”
BackgroundMethodPoseIt: A Visual-Tactile Dataset of Holding Poses for Grasp Stability Analysis
Kanitkar, S., Jiang, H., & Yuan, W. (2022) · IEEE/RSJ International Conference on Intelligent Robots and Systems (IROS) · doi:10.1109/iros47612.2022.9981562 · Cited by 22
“We propose to improve performance on such examples by adapting techniques used for learning with dataset imbalance which under-sample the majority class [10], [33], [34].”
MethodSAFA: Sample-Adaptive Feature Augmentation for Long-Tailed Image Classification
Hong, Y., Zhang, J., Sun, Z., & Yan, K. (2022) · Lecture notes in computer science · doi:10.1007/978-3-031-20053-3_34 · Cited by 22
A comparative study on the effect of data imbalance on software defect prediction
Liu, Y., Zhang, W., Qin, G., & Zhao, J. (2022) · Procedia Computer Science · doi:10.1016/j.procs.2022.11.349 · Cited by 22
Data Augmentation and Classification of Sea–Land Clutter for Over-the-Horizon Radar Using AC-VAEGAN
Zhang, X., Wang, Z., Lu, K., Pan, Q., & Li, Y. (2023) · IEEE Transactions on Geoscience and Remote Sensing · doi:10.1109/tgrs.2023.3274296 · Cited by 22
Bias Mimicking: A Simple Sampling Approach for Bias Mitigation
Qraitem, M., Saenko, K., & Plummer, B. A. (2023) · IEEE Computer Society Conference on Computer Vision and Pattern Recognition/Proceedings - IEEE Computer Society Conference on Computer Vision and Pattern Recognition/Proceedings · doi:10.1109/cvpr52729.2023.01945 · Cited by 22
“Most common of these methods are Undersampling [3, 13, 28] and Oversampling [34].”
BackgroundMethodInfluentialFrom SMOTE to Mixup for Deep Imbalanced Classification
Cheng, W.-C., Mai, T.-H., & Lin, H.-T. (2024) · Communications in computer and information science · doi:10.1007/978-981-97-1711-8_6 · Cited by 25
Intermittent boundary stabilization of stochastic reaction–diffusion Cohen–Grossberg neural networks
Liu, X., Wu, K., & Zhang, W. (2020) · Neural Networks · doi:10.1016/j.neunet.2020.07.019 · Cited by 22
Ning, H., Li, Z., Wang, C., & Yang, L. (2020) · Annals of GIS · doi:10.1080/19475683.2020.1803402 · Cited by 22
“In a systematic study of the class imbalance problem on convolutional neural networks (Buda et al. 2018), the authors report that oversampling outperforms other alternatives in experiments.”
BackgroundDetecting the target of sarcasm is hard: Really??
Parameswaran, P., Trotman, A., Liesaputra, V., & Eyers, D. M. (2021) · Information Processing & Management · doi:10.1016/j.ipm.2021.102599 · Cited by 24
Liu, C., Li, F., Zhang, P., & Balasubramanian, P. (2025) · npj Clean Water · doi:10.1038/s41545-025-00529-4 · Cited by 21
Penalized multiple distribution selection method for imbalanced data classification
Shi, G., Feng, C., Xu, W., Liao, L., & Huang, H. (2020) · Knowledge-Based Systems · doi:10.1016/j.knosys.2020.105833 · Cited by 21
Duarte, D., Andriolo, U., & Gonçalves, G. R. (2020) · ISPRS annals of the photogrammetry, remote sensing and spatial information sciences · doi:10.5194/isprs-annals-v-3-2020-439-2020 · Cited by 21
“…with the literature when addressing the class imbalance problem: oversampling approaches often are the best performing ones (Johnson et al. , 2013; Buda et al. , 2018; Douzas et al. , 2019), even if in this case a binary classification problem was considered and an imbalance ratio of around 1:100.”
BackgroundMethodResultInfluentialMokbal, F. M. M., Wang, D., Wang, X., & Fu, L. (2020) · PeerJ Computer Science · doi:10.7717/peerj-cs.328 · Cited by 21
Saini, M., & Susan, S. (2021) · Multimedia Tools and Applications · doi:10.1007/s11042-021-10612-w · Cited by 24
Liu, J., Guo, F., Gao, H., Huang, Z., Zhang, Y., & Zhou, H. (2021) · Neural Computing and Applications · doi:10.1007/s00521-021-06066-8 · Cited by 23
Classification of Imbalanced Data Using Deep Learning with Adding Noise
Fan, W.-W., & Lee, C. (2021) · Journal of Sensors · doi:10.1155/2021/1735386 · Cited by 23
Artificial intelligence for medical image analysis in epilepsy
Sollee, J., Tang, L., Igiraneza, A. B., Xiao, B., Bai, H. X., & Li, Y. (2022) · Epilepsy Research · doi:10.1016/j.eplepsyres.2022.106861 · Cited by 21
Ma, J. W., & Leite, F. L. (2022) · Automation in Construction · doi:10.1016/j.autcon.2022.104167 · Cited by 21
Impact of loss functions on semantic segmentation in far-field monitoring
Chern, W., Nguyen, T., Asari, V. K., & Kim, H. (2022) · Computer-Aided Civil and Infrastructure Engineering · doi:10.1111/mice.12832 · Cited by 21
“Previous discussions (Buda et al., 2018; He & Garcia, 2009) have already suggested that class imbalance is detrimental to the training process.”
BackgroundMetaGeo: A General Framework for Social User Geolocation Identification With Few-Shot Learning
Zhou, F., Qi, X., Zhang, K., Trajcevski, G., & Zhong, T. (2022) · IEEE Transactions on Neural Networks and Learning Systems · doi:10.1109/tnnls.2022.3154204 · Cited by 21
“Conventional approaches focus on designing data re-sampling strategies, over-sampling the tail classes or down-sampling the head classes [23], cost-sensitive feature representation learning [24], and class-balanced loss designs [25].”
BackgroundSandström, J., Myburgh, H. C., Laurent, C., Swanepoel, D. W., & Lundberg, T. (2022) · Diagnostics · doi:10.3390/diagnostics12061318 · Cited by 24
“It has been shown that oversampling outperforms the other class modification methods and in cases of large class imbalance ratio under-sampling is on par with oversampling [26].”
MethodJiang, C., Lv, W., & Li, J. (2023) · IEEE Transactions on Computational Biology and Bioinformatics · doi:10.1109/tcbb.2023.3238001 · Cited by 21
“Although the existing ML-based prediction methods have made significant progress in PPIs prediction, the severe sample imbalance issue in the PPIs prediction task has a notable negative impact on the effectiveness of ML algorithms [25], [26], [27].”
MethodClass-imbalanced complementary-label learning via weighted loss
Wei, M., Zhou, Y., Li, Z., & Xu, X. (2023) · Neural Networks · doi:10.1016/j.neunet.2023.07.030 · Cited by 22
“The most commonly used strategy is to re-balance the training samples (Buda et al., 2018).”
MethodHostNet: improved sequence representation in deep neural networks for virus-host prediction
Ming, Z., Chen, X., Wang, S., Liu, H. D., Yuan, Z., Wu, M., & Xia, H. (2023) · BMC Bioinformatics · doi:10.1186/s12859-023-05582-9 · Cited by 21
“Additionally, when confronted with imbalanced virus sequence data and data deficiency, deep learning models may need help to learn adequate representations [16], resulting in severe performance deterioration [8].”
BackgroundQiao, W., Shen, L., Wen, Q., Wen, Q., Tang, S., & Li, Z. (2023) · International Journal of Digital Earth · doi:10.1080/17538947.2023.2298245 · Cited by 24
“…due to the reason that the number of collapsed buildings is far less than other objects, three strategies including random over-sampling, random under-sampling, and cost-sensitive were used to deal with the sample imbalance problems (Buda, Maki, and Mazurowski 2018; Ji et al. 2019; Ma et al. 2020).”
ImbSAM: A Closer Look at Sharpness-Aware Minimization in Class-Imbalanced Recognition
Zhou, Y., Qu, Y., Xu, X., & Shen, H. T. (2023) · IEEE/CVF International Conference on Computer Vision (ICCV) · doi:10.1109/iccv51070.2023.01042 · Cited by 26
“The real-world data is usually class-imbalanced [7, 23, 32, 39, 57], following a long-tailed distribution: a”
BackgroundZeng, G., Huang, W., Wang, Y., Wang, X., & Wenjuan, E. (2024) · IEEE Sensors Journal · doi:10.1109/jsen.2024.3360408 · Cited by 23
“Common solutions include data resampling strategy [26], [27], loss reweighting strategy [28], [29], [30], meta-learning [31], and others.”
BackgroundUnsupervised domain adversarial network for few-sample fault detection in industrial processes
Fang, R., Wang, K., Li, J., Yuan, X., & Wang, Y. (2024) · Advanced Engineering Informatics · doi:10.1016/j.aei.2024.102684 · Cited by 21
Deep Discriminative Features Learning and Sampling for Imbalanced Data Problem
Liu, Y.-H., Liu, C., & Tseng, S.-M. (2018) · Proceedings · doi:10.1109/icdm.2018.00150 · Cited by 21
“Moreover, because the original Fashion MNIST dataset is balanced data, we utilized two hyperparameters ρ and μ to create imbalanced data [24].”
MethodProcedural Content Generation of Rhythm Games Using Deep Learning Methods
Yubin, L., Li, W., & Ikeda, K. (2019) · Lecture notes in computer science · doi:10.1007/978-3-030-34644-7_11 · Cited by 21
A Multi-View Deep Evidential Learning Approach for Mammogram Density Classification
Gudhe, N. R., Sudah, M., Sund, R., Kosma, V., Behravan, H., & Mannermaa, A. (2024) · IEEE Access · doi:10.1109/access.2024.3399204 · Cited by 23
“The VinDr-Mammo and DDSM datasets exhibited class imbalance and we addressed this issue using a weighted random sampling approach [60], [61].”
MethodMellit, A., & Kalogirou, S. A. (2025) · Renewable and Sustainable Energy Reviews · doi:10.1016/j.rser.2025.116057 · Cited by 20
Ghribi, F., & Hamdaoui, F. (2025) · The Visual Computer · doi:10.1007/s00371-025-04122-1 · Cited by 24
“For instance, Buda et al. [115] demonstrated that minority-class oversampling combined with a class-balanced cross-entropy loss substantially improves recall on under-representedcategorieswithoutdegradingoverallaccu-racy.”
DeepCXray: Automatically Diagnosing Diseases on Chest X-Rays Using Deep Neural Networks
Xu, X., Guo, Q., Guo, J., & Zhang, Y. (2018) · IEEE Access · doi:10.1109/access.2018.2875406 · Cited by 22
“datasets and the relevant methods can be divided into two categories: data-level and classifier-level [23] approaches.”
MethodArgument Component Classification for Classroom Discussions
Lugini, L., & Litman, D. J. (2018) · Workshop on Argument Mining (ArgMining) · doi:10.18653/v1/w18-5208 · Cited by 21
“To combat this phenomenon we decided to use oversampling (Buda et al., 2017) in order”
MethodA Demand-Driven SAR Target Sample Generation Method for Imbalanced Data Learning
Cao, C., Cui, Z., Wang, L., Wang, J., Cao, Z., & Yang, J. (2021) · IEEE Transactions on Geoscience and Remote Sensing · doi:10.1109/tgrs.2021.3134674 · Cited by 20
ALLIE: Active Learning on Large-scale Imbalanced Graphs
Cui, L.-M., Tang, X.-J., Katariya, S., Rao, N. S., Agrawal, P., Subbian, K., & Lee, D. (2022) · The Web Conference · doi:10.1145/3485447.3512229 · Cited by 28
“A comprehensive review of class imbalance problems in deep learning can be found in [6].”
BackgroundKong, X., Wei, X., Liu, X., Wang, J., Lu, S., Xing, W., & Lu, W. (2022) · Knowledge-Based Systems · doi:10.1016/j.knosys.2022.109561 · Cited by 20
Cross-Hospital Sepsis Early Detection via Semi-Supervised Optimal Transport With Self-Paced Ensemble
Ding, R., Zhou, Y., Xu, J., Xie, Y., Liang, Q., Ren, H., Wang, Y., Chen, Y., Wang, L., & Huang, M. (2023) · IEEE Journal of Biomedical and Health Informatics · doi:10.1109/jbhi.2023.3253208 · Cited by 20
“Imbalanced data is ubiquitous especially for medical diagnostic datasets [20], and it exhibits a long-tailed distribution [21].”
BackgroundEdge-HetIoT defense against DDoS attack using learning techniques
Mahadik, S. S., Pawar, P. M., & Muthalagu, R. (2023) · Computers & Security · doi:10.1016/j.cose.2023.103347 · Cited by 20
Mwaffo, V., Miller, D., & Costello, D. H. (2023) · IEEE Access · doi:10.1109/access.2023.3308822 · Cited by 20
“while neglecting minority classes [36], [37].”
Park, J., Zhong, X., Miley, E. N., Rutledge, R. S., Kakalecik, J., Johnson, M. C., & Gray, C. F. (2023) · Arthroplasty Today · doi:10.1016/j.artd.2023.101308 · Cited by 20
Deep Learning Based Prediction of Pulmonary Hypertension in Newborns Using Echocardiograms
Ragnarsdóttir, H., Özkan, E., Michel, H., Chin-Cheong, K., Manduchi, L., Wellmann, S. M., & Vogt, J. E. (2024) · International Journal of Computer Vision · doi:10.1007/s11263-024-01996-x · Cited by 21
“Strategies to address this imbalance have been extensively studied, and one commonly used approach involves sampling methods (Buda et al., 2018).”
MethodImage classification with category centers in class imbalance situation
Zhang, Y., Shuai, L., Ren, Y., & Chen, H. (2018) · Youth Academic Annual Conference of Chinese Association of Automation (YAC) · doi:10.1109/yac.2018.8406400 · Cited by 20
A Multi-Label Learning Framework for Drug Repurposing
Mei, S., & Zhang, K. (2019) · Pharmaceutics · doi:10.3390/pharmaceutics11090466 · Cited by 23
“Oversampling and undersampling are two major methods to tackle class imbalance in the machine learning field [37].”
Toward Automated Classification of B-Acute Lymphoblastic Leukemia
Kulhalli, R., Savadikar, C., & Garware, B. (2019) · Lecture notes in bioengineering · doi:10.1007/978-981-15-0798-4_7 · Cited by 20
Class-Center Involved Triplet Loss for Skin Disease Classification on Imbalanced Data
Lei, W., Zhang, R., Yang, Y., Wang, R., & Zheng, W. (2020) · IEEE International Symposium on Biomedical Imaging (ISBI) · doi:10.1109/isbi45749.2020.9098718 · Cited by 20
“Besides these traditional approaches, transfer learning by fine-tuning a pre-trained convolutional neural network (CNN) classifier has been proven helpful to improve performance for small-sample classes [7, 8].”
An Efficient Deep Learning Method for Encrypted Traffic Classification on the Web
Soleymanpour, S., Sadr, H., & Beheshti, H. (2020) · International Conference on Web Research (ICWR) · doi:10.1109/icwr49608.2020.9122299 · Cited by 20
“Considering the fact that unbalanced data have a great impact on the performance of CNN and generally cause over-fitting along the raining process [29], various methods have been presented to overcome this issue.”
MethodTowards a Human-AI Hybrid System for Categorising Programming Problems
Pereira, F. D., Pires, F., Fonseca, S. C., Oliveira, E., de Carvalho, L. S. G., de Oliveira, D. B. F., & Cristea, A. I. (2021) · ACM Technical Symposium on Computer Science Education (SIGCSE) · doi:10.1145/3408877.3432422 · Cited by 20
“In these situations, studies [7, 15] suggest adjusting the model by changing the performance metric (e.”
BackgroundA Method Based on Deep Learning for Severe Convective Weather Forecast: CNN-BiLSTM-AM (Version 1.0)
Zhang, J., Yin, M., Wang, P. P., & Gao, Z. (2024) · Atmosphere · doi:10.3390/atmos15101229 · Cited by 21
Schaye, V. E., DiTullio, D. J., Guzman, B. V., Vennemeyer, S., Shih, H., Reinstein, I., Weber, D., Goodman, A., Wu, D. T. Y., Sartori, D. J., et al. (2025) · Journal of Medical Internet Research · doi:10.2196/67967 · Cited by 22
“[41 ] In addition, given these limitations, we did not attempt to fine-tune the 3 separate NYU fine-tuned GatorTron D models at UC.”
Comparing GPT-based approaches in automated writing evaluation
Liu, Y., Lu, X., & Qi, H. (2025) · Assessing Writing · doi:10.1016/j.asw.2025.100961 · Cited by 19
Deep anomaly detection for time series: A survey
Jia, X., Peng, X., Peng, W., Zhao, B., Li, H., & Shen, C. (2025) · Computer Science Review · doi:10.1016/j.cosrev.2025.100787 · Cited by 27
Low-shot transfer with attention for highly imbalanced cursive character recognition
Jalali, A., Kavuri, S., & Lee, M. (2021) · Neural Networks · doi:10.1016/j.neunet.2021.07.003 · Cited by 19
A New Hybrid Under-sampling Approach to Imbalanced Classification Problems
Peng, C.-Y., & Park, Y. (2021) · Applied Artificial Intelligence · doi:10.1080/08839514.2021.1975393 · Cited by 19
A Keyword-Enhanced Approach to Handle Class Imbalance in Clinical Text Classification
Blanchard, A. E., Gao, S. B., Yoon, H., Christian, J. B., Durbin, E. B., Wu, X., Stroup, A. M., Doherty, J. A., Schwartz, S. M., Wiggins, C. L., et al. (2022) · IEEE Journal of Biomedical and Health Informatics · doi:10.1109/jbhi.2022.3141976 · Cited by 19
“to over-fitting and subsequently poor model performance during testing or production [6], [7].”
BackgroundMethodA hierarchical DCNN-based approach for classifying imbalanced water inflow in rock tunnel faces
Chen, J., Huang, H., Cohn, A. G., Zhou, M., Zhang, D., & Man, J. (2022) · Tunnelling and Underground Space Technology · doi:10.1016/j.tust.2022.104399 · Cited by 19
Islam, M. K., Kaushal, C., Amin, M. A., Algarni, A. D., Alturki, N. M., Soliman, N. F., & Mansour, R. F. (2022) · Contrast Media & Molecular Imaging · doi:10.1155/2022/6805460 · Cited by 20
“Furthermore, the metric assist in analysing imbalance datasets more accurately compare to other metrics [109].”
BackgroundChen, T., Chen, J. C., Chang, W., Tsai, W., Shih, M.-C., & Nabila, A. W. (2022) · Journal of Biomedical Informatics · doi:10.1016/j.jbi.2022.104171 · Cited by 19
Jiang, X., Wang, J., Meng, Q., Saada, M., & Cai, H. (2023) · Neural Computing and Applications · doi:10.1007/s00521-023-08290-w · Cited by 20
“Class imbalance is a common problem that has been studied in classical machine learning, yet limited systematic research is available in deep learning [15].”
BackgroundInfluentialDeep learning for network intrusion: A hierarchical approach to reduce false alarms
Moore, S. J., Cruciani, F., Nugent, C., Zhang, S., Cleland, I., & Sani, S. (2023) · Intelligent Systems with Applications · doi:10.1016/j.iswa.2023.200215 · Cited by 23
Margin-Aware Adaptive-Weighted-Loss for Deep Learning Based Imbalanced Data Classification
Roy, D., Pramanik, R., & Sarkar, R. (2023) · IEEE Transactions on Artificial Intelligence · doi:10.1109/tai.2023.3275133 · Cited by 22
Imbalanced Label Distribution Learning
Zhao, X., An, Y., Xu, N., Wang, J., & Geng, X. (2023) · Proceedings of the AAAI Conference on Artificial Intelligence · doi:10.1609/aaai.v37i9.26341 · Cited by 20
“One line of these approaches is re-sampling, which uses under-sampling (Buda, Maki, and Mazurowski 2018) or over-sampling (Byrd and Lipton 2019) to achieve a relatively balanced dataset.”
Method
Metadata
- Authors
- Buda, M., Maki, A., & Mazurowski, M. A.
- Affiliations
- Duke University (US)
- KTH Royal Institute of Technology (SE)
- Corresponding author
- Buda, Mateusz
- Venue
- Neural Networks
- Publisher
- Elsevier BV
- Published
- 2018-10
- Volume
- 106
- Pages
- 249-259
- ISSN
- 0893-6080 (ISSN-L)
- 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
- 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
