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 301–400 of 1,001
Moanna: Multi-Omics Autoencoder-Based Neural Network Algorithm for Predicting Breast Cancer Subtypes
Lupat, R., Perera, R. D., Loi, S., & Li, J. (2023) · IEEE Access · doi:10.1109/access.2023.3240515 · Cited by 77
“Imbalance class training has been studied to affect classifiers’ performance [47], and we hypothesised that this would be one of the reasons for the lower concordance between the predicted HER2-like subtype and the training label.”
BackgroundJia, J. C., & Ye, W. (2023) · Remote Sensing · doi:10.3390/rs15164098 · Cited by 80
“In wireless sensor networks and other relevant technologies deployed in disaster-affected areas, effective disaster response can be achieved by using web crawling or web scraping to obtain data information.”
AREA: Adaptive Reweighting via Effective Area for Long-Tailed Classification
Chen, X., Zhou, Y., Wu, D., Yang, C., Li, B., Hu, Q., & Wang, W. (2023) · IEEE/CVF International Conference on Computer Vision (ICCV) · doi:10.1109/iccv51070.2023.01766 · Cited by 69
“To achieve a balanced training set, we can under-sample the majority classes by randomly discarding some samples, or over-sample the minority classes by duplicating their samples [15, 2, 3].”
MethodConvolutional neural networks as aid in core lithofacies classification
de Lima, R. P., Suriamin, F., Marfurt, K. J., & Pranter, M. J. (2019) · Interpretation · doi:10.1190/int-2018-0245.1 · Cited by 68
“This difference in amount of labeled data for different classes is referred to as class imbalance (Japkowicz and Stephen, 2002; Buda et al., 2018) and can cause undesirable effects when training classifiers.”
BackgroundA multiple combined method for rebalancing medical data with class imbalances
Wang, Y.-C., & Cheng, C.-H. (2021) · Computers in Biology and Medicine · doi:10.1016/j.compbiomed.2021.104527 · Cited by 66
Pan, L.-S., Li, C., Su, S., Tay, S. Y., Tran, Q., & Chan, W. P. (2021) · Scientific Reports · doi:10.1038/s41598-021-93889-z · Cited by 66
An Empirical Evaluation of the t-SNE Algorithm for Data Visualization in Structural Engineering
Hajibabaee, P., Pourkamali‐Anaraki, F., & Hariri‐Ardebili, M. A. (2021) · International Conference on Machine Learning and Applications · doi:10.1109/icmla52953.2021.00267 · Cited by 71
Shamsuddin, I. I. S., Othman, Z., & Sani, N. S. (2022) · Water · doi:10.3390/w14192939 · Cited by 66
“Other metrics, such as accuracy, recall, and F1-score, are frequently used in the research community to evaluate models trained on imbalanced data [37].”
BackgroundLong-Tailed Visual Recognition via Self-Heterogeneous Integration with Knowledge Excavation
Jin, Y., Li, M., Lu, Y., Cheung, Y., & Wang, H. (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/cvpr52729.2023.02269 · Cited by 70
“The most straightforward action for long-tailed recognition often focuses on re-balancing the learning process from either a data processing [3, 27, 46] or cost-sensitive perspective [12, 28, 71].”
BackgroundChen, X., Liu, C., Chen, L., Zhu, X., Zhang, Y., & Wang, C. (2024) · Applied Sciences · doi:10.3390/app14031157 · Cited by 66
“Generally, the problem of imbalanced sample distribution in datasets can often lead to overfitting of the model [30].”
Zaghloul, M., Barakat, S., & Rezk, A. (2024) · Journal of Retailing and Consumer Services · doi:10.1016/j.jretconser.2024.103865 · Cited by 66
A review of deep learning methods for digitisation of complex documents and engineering diagrams
Jamieson, L., Moreno‐García, C. F., & Elyan, E. (2024) · Artificial Intelligence Review · doi:10.1007/s10462-024-10779-2 · Cited by 78
“The problem of class imbalance is known to occur in both deep learning and traditional machine learning (Buda et al. 2018).”
The Integration of Radiomics and Artificial Intelligence in Modern Medicine
Maniaci, A., Lavalle, S., Gagliano, C., Lentini, M., Masiello, E., Parisi, F. M., Iannella, G., Cilia, N. D., Salerno, V. M., Cusumano, G., et al. (2024) · Life · doi:10.3390/life14101248 · Cited by 70
“In many situations, these models have outperformed conventional machine learning techniques in tasks including tumor segmentation, lesion identification, and image classification [21,22].”
MethodGenerative Oversampling Method for Imbalanced Data on Bearing Fault Detection and Diagnosis
Suh, S., Lee, H., Jo, J. H., Lukowicz, P., & Lee, Y. O. (2019) · Applied Sciences · doi:10.3390/app9040746 · Cited by 66
“Additionally, oversampling is always the most effective way to deal with class imbalance for the CNN model on image classification [24].”
MethodLearning to Amend Facial Expression Representation via De-albino and AffinityPreprint
Shi, J., Zhu, S., & Liang, Z. (2021) · arXiv (Cornell University) · doi:10.48550/arxiv.2103.10189 · Cited by 82
Image-based classification of plant genus and family for trained and untrained plant species
Seeland, M., Rzanny, M., Boho, D., Wäldchen, J., & Mäder, P. (2019) · BMC Bioinformatics · doi:10.1186/s12859-018-2474-x · Cited by 64
“Imbalance can cause low accuracy on under-represented classes [17].”
BackgroundActive, continual fine tuning of convolutional neural networks for reducing annotation efforts
Zhou, Z., Shin, J. Y., Gurudu, S., Gotway, M. B., & Liang, J. (2021) · Medical Image Analysis · doi:10.1016/j.media.2021.101997 · Cited by 74
Lobert, F., Holtgrave, A.-K., Schwieder, M., Pause, M., Vogt, J., Gocht, A., & Erasmi, S. (2021) · Remote Sensing of Environment · doi:10.1016/j.rse.2021.112751 · Cited by 61
“Even if the evaluation of a particular classification algorithm was not the main aspect, our work offered insights into the classification power of satellite time series with 1D CNNs and provides a basis for further evaluation regarding both the optimization of classification algorithms for dense satellite time series…”
InfluentialKim, K., & Sohn, S. Y. (2020) · Neural Networks · doi:10.1016/j.neunet.2020.06.026 · Cited by 63
Machine Learning Methods for Preterm Birth Prediction: A Review
Wlodarczyk, T. W., Płotka, S., Szczepański, T., Rokita, P. S., Sochacki-Wójcicka, N., Wójcicki, J., Lipa, M., & Trzcinski, T. P. (2021) · Electronics · doi:10.3390/electronics10050586 · Cited by 78
“However, some methods for coping with the imbalanced class in Convolutional Neural Networks are reviewed by Buda et al. [28].”
MethodHan, S.-Y., Zhong, X. Y., Shao, H., Xu, T., Zhao, R., & Cheng, J. (2021) · Measurement Science and Technology · doi:10.1088/1361-6501/ac1b43 · Cited by 62
“Under this situation, CNN tends to incline to the large samples, which will result in performance degradation [29].”
Hashemi, M., Akhbari, M., & Jutten, C. (2022) · Computers in Biology and Medicine · doi:10.1016/j.compbiomed.2022.105402 · Cited by 63
A hybrid deep learning skin cancer prediction framework
Farea, E., Saleh, R. A. A., AbuAlkebash, H., Farea, A. A. R., & Al–antari, M. A. (2024) · Engineering Science and Technology an International Journal · doi:10.1016/j.jestch.2024.101818 · Cited by 62
Adjusting Decision Boundary for Class Imbalanced Learning
Kim, B., & Kim, J. (2020) · IEEE Access · doi:10.1109/access.2020.2991231 · Cited by 72
“In addition, a step imbalance [2] was implanted to verify whether our proposed method can resolve various types of imbalance.”
BackgroundMethodFLEX: Unifying Evaluation for Few-Shot NLPPreprint
Bragg, J., Cohan, A., Lo, K., & Beltagy, I. (2021) · arXiv (Cornell University) · doi:10.48550/arxiv.2107.07170 · Cited by 69
“Class imbalance has also been observed to degrade performance [11, 47], yet prior few-shot NLP evaluations do not include this setting either.”
BackgroundFINE Samples for Learning with Noisy LabelsPreprint
Kim, T., Ko, J., Cho, S., Choi, J., & Yun, S.-Y. (2021) · arXiv (Cornell University) · doi:10.48550/arxiv.2102.11628 · Cited by 62
“In class-imbalance tasks, numerous studies have conducted over-sampling of minority classes [7, 1] or undersampling the majority classes [5] to balance the amount of data per class.”
BackgroundAUC-Based Extreme Learning Machines for Supervised and Semi-Supervised Imbalanced Classification
Wang, G., Wong, K. W., & Lü, J. (2020) · IEEE Transactions on Systems Man and Cybernetics Systems · doi:10.1109/tsmc.2020.2982226 · Cited by 61
“First, class imbalance is a common challenge in practice, such as cancer detection [45], computer vision [46], and fraud detection [47], where one class has a much larger number of training samples than the other class.”
BackgroundFeature construction and smote-based imbalance handling for multi-label learning
Mishra, N. K., & Singh, P. K. (2021) · Information Sciences · doi:10.1016/j.ins.2021.03.001 · Cited by 61
Automated Endoscopic Image Classification via Deep Neural Network With Class Imbalance Loss
Yue, G., Wei, P., Liu, Y., Luo, Y., Du, J., & Wang, T. (2023) · IEEE Transactions on Instrumentation and Measurement · doi:10.1109/tim.2023.3264047 · Cited by 61
“In the literature, the algorithmic-level methods, including ensemble-based and cost-sensitive ones, mainly focus on adjusting the classifiers and have become good choices to deal with the imbalance problems [44].”
BackgroundGradient Reweighting: Towards Imbalanced Class-Incremental Learning
He, J. (2024) · 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/cvpr52733.2024.01577 · Cited by 70
“Specifically, the re-sampling based methods [8, 35, 43] aim to construct a balanced training batch by increasing the probability of tail classes to be sampled.”
MethodCardiomegaly Detection on Chest Radiographs: Segmentation Versus Classification
Sogancioglu, E., Murphy, K., Çallı, E., Scholten, E. T., Schalekamp, S., & van Ginneken, B. (2020) · IEEE Access · doi:10.1109/access.2020.2995567 · Cited by 64
“Since there is a class imbalance problem in such a scenario (1156 images with cardiomegaly among 65k), we employed an over-sampling technique [4] by sampling the positive cardiomegaly cases until the dataset was balanced.”
MethodDeep learning to map concentrated animal feeding operations
Handan‐Nader, C., & Ho, D. E. (2019) · Nature Sustainability · doi:10.1038/s41893-019-0246-x · Cited by 60
Chen, X., Zhou, Y., Wu, D., Zhang, W., Zhou, Y., Li, B., & Wang, W. (2022) · Proceedings of the AAAI Conference on Artificial Intelligence · doi:10.1609/aaai.v36i1.19912 · Cited by 60
Uncertainty-Aware Learning against Label Noise on Imbalanced Datasets
Huang, Y., Bai, B., Zhao, S., Bai, K., & Wang, F. (2022) · Proceedings of the AAAI Conference on Artificial Intelligence · doi:10.1609/aaai.v36i6.20654 · Cited by 62
“Another common problem in real-life applications is that some classes have a higher volume of examples in the training set than other classes, which we refer to as the class imbalance problem (Buda, Maki, and Mazurowski 2018).”
BackgroundA Comparison of Techniques for Class Imbalance in Deep Learning Classification of Breast Cancer
Walsh, R., & Tardy, M. (2022) · Diagnostics · doi:10.3390/diagnostics13010067 · Cited by 69
“This problem of different proportions of each class of interest is called class imbalance, and numerous studies have shown that it can be detrimental to the performance of a classification model [7,8].”
BackgroundResultInfluentialContract-Theory-Based Incentive Mechanism for Federated Learning in Health CrowdSensing
Li, L., Yu, X., Cai, X., He, X., & Liu, Y. (2022) · IEEE Internet of Things Journal · doi:10.1109/jiot.2022.3218008 · Cited by 63
“In the establishment of health monitoring and disease prediction models, the data of a single hospital or community medical center is not enough to train the model, so multiple organizations need to cooperate to train the model for extract crowd intelligence [15], [16].”
BackgroundElujide, I., Fashoto, S. G., Fashoto, B., Mbunge, E., Folorunso, S. O., & Olamijuwon, J. O. (2021) · Informatics in Medicine Unlocked · doi:10.1016/j.imu.2021.100545 · Cited by 67
Binary imbalanced data classification based on diversity oversampling by generative models
Zhai, J., Qi, J., & Shen, C. (2021) · Information Sciences · doi:10.1016/j.ins.2021.11.058 · Cited by 59
Cao, Y. J., Xu, H., Su, C., & Yang, Q. (2022) · IEEE Transactions on Power Delivery · doi:10.1109/tpwrd.2022.3202958 · Cited by 60
“As suggested in [7], the appropriate resampling may not lead to the overfitting of the mainstream neural networks.”
BackgroundEpileptic Disorder Detection of Seizures Using EEG Signals
Alharthi, M. K., Moria, K. M., Alghazzawi, D., & Tayeb, H. O. (2022) · Sensors · doi:10.3390/s22176592 · Cited by 59
“This imbalanced dataset issue can have a major negative impact on classification performance [10].”
BackgroundRadiomics Applications in Head and Neck Tumor Imaging: A Narrative Review
Tortora, M., Gemini, L., Scaravilli, A., Ugga, L., Ponsiglione, A., Stanzione, A., D’Arco, F., D’Anna, G., & Cuocolo, R. (2023) · Cancers · doi:10.3390/cancers15041174 · Cited by 59
“The risk of overfitting is reduced by oversampling and undersampling, which prevent the model from seeing significantly more instances of one class than others during training [113].”
BackgroundChungath, T. T., Nambiar, A., & Mittal, A. (2023) · IEEE Journal of Oceanic Engineering · doi:10.1109/joe.2022.3221127 · Cited by 62
“The literature suggests data level and classifier level methods to address this issue [32].”
BackgroundXiao, X., Ming, W., Luo, X., Yang, L., Li, M., Yang, P., Ji, X., & Li, Y. (2024) · Agricultural Water Management · doi:10.1016/j.agwat.2024.108692 · Cited by 59
Deep Attention-Based Imbalanced Image Classification
Wang, L., Zhang, L., Qi, X., & Zhang, Y. (2021) · IEEE Transactions on Neural Networks and Learning Systems · doi:10.1109/tnnls.2021.3051721 · Cited by 61
CoSSL: Co-Learning of Representation and Classifier for Imbalanced Semi-Supervised Learning
Fan, Y., Dai, D., Kukleva, A., & Schiele, B. (2022) · 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/cvpr52688.2022.01417 · Cited by 63
“nition [6, 9, 19, 20, 24] is not designed to utilize unlabeled data despite being good at handling data imbalance.”
BackgroundMancardi, G., Mikołajczyk, A., Annapoorani, V. K., Bahl, A., Blekos, K., Burk, J., Çetin, Y. A., Chairetakis, K., Dutta, S., Escorihuela, L., et al. (2023) · Materials Today · doi:10.1016/j.mattod.2023.05.029 · Cited by 60
VulExplainer: A Transformer-Based Hierarchical Distillation for Explaining Vulnerability Types
Fu, M. C., Nguyen, V., Tantithamthavorn, C. K., Le, T., & Phung, D. (2023) · IEEE Transactions on Software Engineering · doi:10.1109/tse.2023.3305244 · Cited by 62
“Furthermore, Zhang et al. [78] explored prevalent re-sampling methods [7], [34], [48] and data augmentation techniques [70], [77] for effectively addressing long-tailed data challenges within the domain of visual recognition.”
BackgroundPouliot, D., Latifovic, R., Pasher, J., & Duffe, J. A. (2019) · Remote Sensing · doi:10.3390/rs11070772 · Cited by 58
“For deep leaning, class imbalance ratios typically much greater than 10:1 have been shown to negatively impact accuracy, but it is related to the specific classification task, number of minority classes, and the class distribution structure [51].”
BackgroundLuo, Z., Zeng, X., Bao, Z., & Xu, M. (2019) · International Joint Conference on Neural Networks (IJCNN) · doi:10.1109/ijcnn.2019.8851972 · Cited by 57
“Experiments have also been done to compare the multipath CNN with traditional classifier, Granular Support Vector Machine with Repetitive Undersampling(GSVM-RU), [47], [48].”
MethodFully Automated Segmentation of Head CT Neuroanatomy Using Deep Learning
Cai, J. C., Akkus, Z., Philbrick, K. A., Boonrod, A., Hoodeshenas, S., Weston, A. D., Rouzrokh, P., Conte, G. M., Zeinoddini, A., Vogelsang, D. C., et al. (2020) · Radiology Artificial Intelligence · doi:10.1148/ryai.2020190183 · Cited by 59
“Class imbalance degrades CNN performance, as models that learn from imbalanced datasets are biased to reflect the frequency of training samples (33).”
BackgroundThiagarajan, P., Khairnar, P., & Ghosh, S. (2021) · IEEE Transactions on Medical Imaging · doi:10.1109/tmi.2021.3123300 · Cited by 68
“CNN is known to perform poorly for biased data set [62].”
BackgroundDeep learning accelerated prediction of the permeability of fibrous microstructures
Çağlar, B., Broggi, G., Ali, M. A., Orgéas, L., & Michaud, V. J. (2022) · Composites Part A Applied Science and Manufacturing · doi:10.1016/j.compositesa.2022.106973 · Cited by 62
VL-LTR: Learning Class-wise Visual-Linguistic Representation for Long-Tailed Visual Recognition
Tian, C., Wang, W., Zhu, X., Dai, J., & Qiao, Y. (2022) · Lecture notes in computer science · doi:10.1007/978-3-031-19806-9_5 · Cited by 65
“One type of the class re-balanced strategy is Data Resampling [1,3,5,11,13,40], which generates class-balanced data by adjusting the sampling rate of tail classes and head classes, yet they might take the risk of over-fitting on datascarced classes.”
BackgroundUnbiased Subclass Regularization for Semi-Supervised Semantic Segmentation
Guan, D., Huang, J., Xiao, A., & Lu, S. (2022) · 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/cvpr52688.2022.00973 · Cited by 60
“Note the class imbalance issue has been widely studied in supervised learning via re-sampling [5, 6, 34, 55, 62], ar X iv :2 20 3.”
BackgroundMethodHenning, S., Beluch, W., Fraser, A., & Friedrich, A. (2023) · Conference of the European Chapter of the Association for Computational Linguistics · doi:10.18653/v1/2023.eacl-main.38 · Cited by 72
Kim, J.-K., & Kim, Y. (2018) · Interspeech (USB) · doi:10.21437/interspeech.2018-1581 · Cited by 58
Zhou, J., Dai, H., & Wang, H. H. (2019) · ACM Transactions on Intelligent Systems and Technology · doi:10.1145/3339308 · Cited by 56
“In general, raw datasets often have the imbalance class problem [Buda et al. 2018], since the origin data is always raw and crude during experiments or productions.”
BackgroundThakkar, H. K., Liao, W.-W., Wu, C., Hsieh, Y., & Lee, T. (2020) · Journal of NeuroEngineering and Rehabilitation · doi:10.1186/s12984-020-00758-3 · Cited by 56
DiagSet: a dataset for prostate cancer histopathological image classification
Koziarski, M., Cyganek, B., Niedziela, P., Olborski, B., Antosz, Z., Żydak, M., Kwolek, B., Wąsowicz, P., Bukała, A., Swadźba, J., et al. (2024) · Scientific Reports · doi:10.1038/s41598-024-52183-4 · Cited by 67
“Data imbalance is known to negatively a ff ect the performance of various classification algorithms, including convolutional neural networks (Buda et al., 2018).”
BackgroundDangut, M. D., Jennions, I. K., King, S., & Skaf, Z. (2022) · Neural Computing and Applications · doi:10.1007/s00521-022-07167-8 · Cited by 67
“Many researchers have agreed that the subject of imbalanced data with deep learning is understudied [21–24].”
BackgroundPerez, A. G., Abd-Elrahman, A. H., Wilkinson, B., Johnson, D. J., & Carthy, R. R. (2022) · Remote Sensing · doi:10.3390/rs14163937 · Cited by 58
Shao, J., Qu, C., Li, J., & Peng, S. (2018) · Sensors · doi:10.3390/s18093039 · Cited by 56
“[47] investigated the impact of class imbalance on the classification performance of CNNs and compared some frequently used methods.”
MethodDemogPairs: Quantifying the Impact of Demographic Imbalance in Deep Face Recognition
Hupont, I., & Fernández, C. (2019) · International Conference on Automatic Face and Gesture Recognition · doi:10.1109/fg.2019.8756625 · Cited by 56
“imbalanced at several levels, which may dramatically impact the performance of deep models [13], [2].”
BackgroundIntrusion Detection for Smart Home Security Based on Data Augmentation with Edge Computing
Yuan, D., Ota, K., Dong, M., Zhu, X., Wu, T., Zhang, L., & Ma, J. (2020) · International Conference on Communications · doi:10.1109/icc40277.2020.9148632 · Cited by 56
“When the training process encountered the problem of imbalanced data, the performance of CNN could turn to deterioration [17].”
MethodResolving Class Imbalance in Object Detection with Weighted Cross Entropy LossesPreprint
Phan, T. H., & Yamamoto, K. (2020) · arXiv (Cornell University) · doi:10.48550/arxiv.2006.01413 · Cited by 89
“…existing data-level methods to tackle the problem such as over-sampling of minority object classes or under-sampling of majority object classes (M. Buda et al., 2018), in this paper, we focus ourselves to cost-sensitive approach by simply applying different weighted versions of Cross Entropy…”
BackgroundMethodInfluentialABC: Auxiliary Balanced Classifier for Class-imbalanced Semi-supervised LearningPreprint
Lee, H., Shin, S., & Kim, H. (2021) · arXiv (Cornell University) · doi:10.48550/arxiv.2110.10368 · Cited by 57
“We also consider step imbalance [5] in which the whole majority classes have the same amount of data and the whole minority classes also have the same amount of data.”
BackgroundBreast cancer histopathology image classification using kernelized weighted extreme learning machine
Saxena, S., Shukla, S., & Gyanchandani, M. (2020) · International Journal of Imaging Systems and Technology · doi:10.1002/ima.22465 · Cited by 57
“when it has a significantly higher number of samples of one class as compared to the other class.(7) The class whose number of samples is more than the average number of samples per class is called the majority type, and the class whose number of samples is less than the average number of samples per class is called t…”
BackgroundFew-Shot Learning With Class Imbalance
Ochal, M., Patacchiola, M., Vazquez, J., Storkey, A., & Wang, S. (2023) · IEEE Transactions on Artificial Intelligence · doi:10.1109/tai.2023.3298303 · Cited by 69
“The weighted loss [7], [8] defines a set of weights for each class, giving more or less weight based on the number of samples associated to the class.”
BackgroundMethodInfluentialAli, M. A., Guan, Q., Umer, R., Cantwell, W. J., & Zhang, T. (2021) · Composites Science and Technology · doi:10.1016/j.compscitech.2021.109091 · Cited by 55
Afify, H. M., Mohammed, K. K., & Hassanien, A. E. (2023) · Biomedical Signal Processing and Control · doi:10.1016/j.bspc.2023.104704 · Cited by 55
Rethinking Importance Weighting for Deep Learning under Distribution ShiftPreprint
Fang, T., Lu, N., Niu, G., & Sugiyama, M. (2020) · arXiv (Cornell University) · doi:10.48550/arxiv.2006.04662 · Cited by 69
“Class-prior-shift experiments We impose class-prior shift on Fashion-MNIST following [4]: • the classes are divided into majority classes and minority classes, where the fraction of the minority classes is μ < 1; • the training data are drawn from every majority class using a sample size, and from every minority class…”
BackgroundMethodInfluentialUsing Improved Conditional Generative Adversarial Networks to Detect Social Bots on Twitter
Wu, B., Liu, L., Yang, Y., Zheng, K., & Wang, X. (2020) · IEEE Access · doi:10.1109/access.2020.2975630 · Cited by 54
“These approaches can be broadly divided into two main categories: data-driven approaches and algorithm-driven approaches [35], [36].”
BackgroundDong, Y., Shen, X., Jiang, Z., & Wang, H. (2020) · Applied Acoustics · doi:10.1016/j.apacoust.2020.107740 · Cited by 55
Rani, M., & Gagandeep, G. (2022) · Multimedia Tools and Applications · doi:10.1007/s11042-021-11747-6 · Cited by 56
“In this approach, the penalty for the misclassification of minority classes is applied to force the learning algorithms to focus more on these classes [8].”
BackgroundMethodHepatitis C Virus Detection Model by Using Random Forest, Logistic-Regression and ABC Algorithm
Li, T. S., Chiu, H.-J., & Kuo, P. (2022) · IEEE Access · doi:10.1109/access.2022.3202295 · Cited by 54
Harfouche, A. L., Nakhle, F., Harfouche, A., Sardella, O. G., Dart, E., & Jacobson, D. (2022) · Trends in Plant Science · doi:10.1016/j.tplants.2022.08.021 · Cited by 56
On Multi-Domain Long-Tailed Recognition, Imbalanced Domain Generalization and Beyond
Yang, Y., Wang, H., & Katabi, D. (2022) · Lecture notes in computer science · doi:10.1007/978-3-031-20044-1_4 · Cited by 61
“, instead of a uniform label distribution over classes, in reality, data are by their nature imbalanced: a few classes contain a large number of instances, whereas many others have only a few instances [5, 6, 52].”
BackgroundThreshold optimization and random undersampling for imbalanced credit card data
Leevy, J. L., Johnson, J. M., Hancock, J., & Khoshgoftaar, T. M. (2023) · Journal Of Big Data · doi:10.1186/s40537-023-00738-z · Cited by 54
“Buda et al. [13] investigated the effect of output thresholding on Convolutional Neural Networks (CNNs) [14], with the aid of three benchmark datasets: MNIST [15], CIFAR-10 [16], and ImageNet [17].”
MethodIslam, M. M., Alam, M. J., Maniruzzaman, M., Ahmed, N. A. M. F., Ali, M. S., Rahman, M. J., & Roy, D. C. (2023) · PLoS ONE · doi:10.1371/journal.pone.0289613 · Cited by 54
Zhao, J., Yu, Y., Lu, S., Huang, Y., Yu, X., Mao, X., Yu, Y., Li, Z., & Liu, H. (2025) · Trends in Food Science & Technology · doi:10.1016/j.tifs.2025.105162 · Cited by 54
Yin, C., Zeng, T., Zhang, H., Fu, W., Wang, L., & Yao, S. (2022) · Agronomy · doi:10.3390/agronomy12040906 · Cited by 54
“As shown in Figure 3e, more than 1/2 of the leaves have wilting symptoms when the maize small leaf spot disease is grade 4, which is quite different from other grades.”
BackgroundModel-Based Synthetic Geoelectric Sampling for Magnetotelluric Inversion With Deep Neural Networks
Li, R., Yu, N., Wang, X., Liu, Y., Cai, Z., & Wang, E. (2020) · IEEE Transactions on Geoscience and Remote Sensing · doi:10.1109/tgrs.2020.3043661 · Cited by 53
“Methods of handling imbalanced data sets can be used to improve the distribution of the samples to avoid overfitting; this is necessary because most standard machine learning methods assume or expect balanced class distributions or equal misclassification costs [29], [30].”
MethodRevisiting Distillation and Incremental Classifier Learning
Javed, K., & Shafait, F. (2019) · Lecture Notes in Computer Science · doi:10.1007/978-3-030-20876-9_1 · Cited by 53
“[3] showed that threshold moving is effective in removing bias in ANNs.”
BackgroundDoes Head Label Help for Long-Tailed Multi-Label Text Classification
Xiao, L., Zhang, X., Jing, L., Huang, C., & Song, M. (2021) · Proceedings of the AAAI Conference on Artificial Intelligence · doi:10.1609/aaai.v35i16.17660 · Cited by 57
“…strategy : The most popular ideas include under-sampling the head classes (Byrd and Lipton 2019), over-sampling the tail classes (Chawla et al. 2002; Buda, Maki, and Mazurowski 2018; Byrd and Lip-ton 2019), and allocating large weights to tail classes in loss functions (Cui et al. 2019).”
BackgroundWeakly-Supervised Network for Detection of COVID-19 in Chest CT Scans
Mohammed, A. S., Wang, C., Zhao, M. Y., Ullah, M., Naseem, R., Wang, H. H., Pedersen, M., & Cheikh, F. A. (2020) · IEEE Access · doi:10.1109/access.2020.3018498 · Cited by 52
“Class balancing strategies including resampling, oversampling, and cost-sensitive reweighting have been incorporated in several classification methods to resolve the class imbalance challenge and considerably improve the performance of an algorithm [76].”
MethodSultan, H., Owais, M., Park, C., Mahmood, T., Haider, A., & Park, K.-R. (2021) · Journal of Personalized Medicine · doi:10.3390/jpm11060482 · Cited by 53
“These metrics are commonly used to evaluate classification frameworks [65] and are calculated as follows:”
MethodConvolutional Fine-Grained Classification With Self-Supervised Target Relation Regularization
Liu, K., Chen, K., & Jia, K. (2022) · IEEE Transactions on Image Processing · doi:10.1109/tip.2022.3197931 · Cited by 54
A Survey on Unbalanced Classification: How Can Evolutionary Computation Help?
Pei, W., Xue, B., Zhang, M., Shang, L., Yao, X., & Zhang, Q. (2023) · IEEE Transactions on Evolutionary Computation · doi:10.1109/tevc.2023.3257230 · Cited by 56
“…several survey papers to discuss various classification algorithms with different techniques to address the class imbalance issue [4], [20], [21], [22], [23], [24], [25], [26], [27], [28], [29], [30], [31], [32], [33], but none of them has specifically and systemically investigated how EC can be…”
BackgroundBenkendorf, D. J., Schwartz, S. D., Cutler, D. R., & Hawkins, C. P. (2023) · Ecological Modelling · doi:10.1016/j.ecolmodel.2023.110414 · Cited by 61
Data Augmentation for Object Detection via Progressive and Selective Instance-SwitchingPreprint
Wang, H., Wang, Q., Yang, F., Zhang, W., & Zuo, W. (2019) · arXiv (Cornell University) · doi:10.48550/arxiv.1906.00358 · Cited by 61
“Among them, methods [3, 4, 22] balance distribution of classes by re-sampling for those containing less training samples.”
BackgroundMethodIdentifying and Compensating for Feature Deviation in Imbalanced Deep LearningPreprint
Ye, H.-J., Chen, H.-Y., Zhan, D., & Chao, W. (2020) · arXiv (Cornell University) · doi:10.48550/arxiv.2001.01385 · Cited by 98
“, mini-batch SGD on uniformly sampled instances) have been found to perform poorly on minor classes [3, 19, 40, 52], which is particularly unfavorable if we evaluate the classifiers with class-balanced test data or average per-class accuracy.”
BackgroundMethodInfluentialLiu, H., Iwamoto, N., Zhu, Z., Li, Z., Zhou, Y., Bozkurt, E., & Zheng, B. (2022) · ACM International Conference on Multimedia (ACM MM) · doi:10.1145/3503161.3548400 · Cited by 55
“The basic idea of resampling-based methods is to over-sample the minority categories [18, 30] or to under-sample the frequent categories during the training process [13, 23].”
MethodA Re-Balancing Strategy for Class-Imbalanced Classification Based on Instance Difficulty
Yu, S., Guo, J., Zhang, R., Fan, Y., Wang, Z., & Cheng, X. (2022) · 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/cvpr52688.2022.00017 · Cited by 50
“However, the datasets in real-world applications often exhibit imbalanced data distributions [27,30].”
BackgroundMethodLi, X., Zhao, D., Xie, J., Wen, H., Liu, C., Li, Y., Li, W., & Wang, S. (2023) · BMC Oral Health · doi:10.1186/s12903-023-03751-z · Cited by 62
“Random shift augmentation, oversampling, adjusting weights in the loss function, and transfer learning were used to overcome class-imbal-anced issues, which detrimentally contributed to DL classification performance [30, 39, 41, 42, 50, 51, 53].”
MethodGAN-based image-to-friction generation for tactile simulation of fabric material
Cai, S., Zhao, L., Ban, Y., Narumi, T., Liu, Y., & Zhu, K. (2021) · Computers & Graphics · doi:10.1016/j.cag.2021.09.007 · Cited by 50
“The problem of data imbalance may further bias the predictive/generative capability of the deep neural network, leading to the worse performance on the minority class than the other classes [67].”
BackgroundExtracting keyframes of breast ultrasound video using deep reinforcement learning
Huang, R., Ying, Q., Lin, Z., Zheng, Z., Tan, L., Tang, G., Zhang, Q., Luo, M., Yi, X., Liu, P., et al. (2022) · Medical Image Analysis · doi:10.1016/j.media.2022.102490 · Cited by 50
Comprehensive machine and deep learning analysis of sensor-based human activity recognition
Balaha, H. M., & Hassan, A. E.-S. (2023) · Neural Computing and Applications · doi:10.1007/s00521-023-08374-7 · Cited by 52
“Different methods were proposed to deal with the class imbalance problem and these methods can be split into two main approaches: data-level and classifier-level methods [35].”
MethodWu, C., Huang, H., Zhang, L., Chen, J., Tong, Y., & Zhou, M. (2023) · Tunnelling and Underground Space Technology · doi:10.1016/j.tust.2023.105432 · Cited by 50
Source-Free Unsupervised Domain Adaptation: Current research and future directions
Zhang, N., Lü, J., Li, K., Fang, Z., & Zhang, G. (2023) · Neurocomputing · doi:10.1016/j.neucom.2023.126921 · Cited by 52
A novel methodology to classify test cases using natural language processing and imbalanced learning
Tahvili, S., Hatvani, L., Ramentol, E., Pimentel, R., Afzal, W., & Herrera, F. (2020) · Engineering Applications of Artificial Intelligence · doi:10.1016/j.engappai.2020.103878 · Cited by 50
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
