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 501–600 of 1,001
Otsuka, K., & Tsumori, M. (2020) · IEEE Access · doi:10.1109/access.2020.3041672 · Cited by 43
Balancing Approaches towards ML for IDS: A Survey for the CSE-CIC IDS Dataset
Gopalan, S. S., Ravikumar, D., Linekar, D., Raza, A., & Hasib, M. (2021) · International Conference on Communications, Signal Processing, and their Applications (ICCSPA) · doi:10.1109/iccspa49915.2021.9385742 · Cited by 38
“In [10] and [11] the authors have reported on the accuracy of proposed ML techniques which account for the imbalance of the dataset.”
BackgroundSequential targeting: A continual learning approach for data imbalance in text classification
Jang, J., Kim, Y., Choi, K.-H., & Suh, S. (2021) · Expert Systems with Applications · doi:10.1016/j.eswa.2021.115067 · Cited by 44
“Following the conventions of the previous research on imbalanced data (Buda et al., 2018; Johnson & Khoshgoftaar, 2019), we employ two distinct ratios used throughout the experiment to represent the imbalanced state of the data.”
BackgroundMethodSelf-supervised ECG pre-training
Liu, H., Zhao, Z., & She, Q. (2021) · Biomedical Signal Processing and Control · doi:10.1016/j.bspc.2021.103010 · Cited by 39
MosaicOS: A Simple and Effective Use of Object-Centric Images for Long-Tailed Object Detection
Zhang, C., Pan, T.-Y., Li, Y., Hu, H., Xuan, D., Changpinyo, S., Gong, B., & Chao, W. (2021) · IEEE/CVF International Conference on Computer Vision (ICCV) · doi:10.1109/iccv48922.2021.00047 · Cited by 41
“The rational is that a classifier trained with long-tailed data tends to assign lower probabilities to minor classes of scarce training data [6, 38, 39, 84].”
Development of a convolutional neural network to detect abdominal aortic aneurysms
Camara, J. R., Tomihama, R. T., Pop, A., Shedd, M. P., Dobrowsky, B., Knox, C. J., Abou‐Zamzam, A. M., & Kiang, S. C. (2022) · Journal of Vascular Surgery Cases and Innovative Techniques · doi:10.1016/j.jvscit.2022.04.003 · Cited by 39
“21,22 Model development and analysis were performed using Keras (version 2.4.”
MethodInfluentialRoy, S., Tyagi, M., Bansal, V., & Jain, V. (2022) · Computers in Biology and Medicine · doi:10.1016/j.compbiomed.2022.106092 · Cited by 42
“The problem of class imbalance appears when the number of images in one class outnumbered the other classes [10] and consequently it may affect the final results of the classification.”
Feng, Q., Xu, P., Ma, D., Lan, G., Wang, F., Wang, D., & Yun, Y. (2022) · Precision Agriculture · doi:10.1007/s11119-022-09959-3 · Cited by 37
“For this issue, Gao et al., (2021) proposed a data balance algorithm (Buda et al., 2018 ; Johnson & Khoshgoftaar, 2019) and applied the algorithm to research on the quality classification and identification of pistachios.”
BackgroundMethodLAN: Learning Adaptive Neighbors for Real-Time Insider Threat Detection
Cai, X., Wang, Y., Xu, S., Li, H., Zhang, Y., Liu, Z., & Yuan, X. (2024) · IEEE Transactions on Information Forensics and Security · doi:10.1109/tifs.2024.3488527 · Cited by 46
“To clearly show the data imbalance problem, we calculate the imbalance ratio (IR) by N maj / N min , where N maj and N min are the sample sizes of the majority class and the minority class, respectively [33].”
MethodAdaptive Subspace Optimization Ensemble Method for High-Dimensional Imbalanced Data Classification
Xu, Y., Yu, Z., Chen, C. L. P., & Liu, Z. (2021) · IEEE Transactions on Neural Networks and Learning Systems · doi:10.1109/tnnls.2021.3106306 · Cited by 39
“Buda et al. [78] improve performance of convolutional neural networks (CNNs) for skewed data classification by integrating different resampling methods.”
BackgroundTOPSIS aided ensemble of CNN models for screening COVID-19 in chest X-ray images
Pramanik, R., Dey, S., Malakar, S., Mirjalili, S., & Sarkar, R. (2022) · Scientific Reports · doi:10.1038/s41598-022-18463-7 · Cited by 38
Chen, J., Liu, Y., & Hou, J.-A. (2023) · Structural Health Monitoring · doi:10.1177/14759217221139730 · Cited by 37
Adaptive Risk Minimization: A Meta-Learning Approach for Tackling Group ShiftPreprint
Zhang, M. M., Marklund, H., Dhawan, N., Gupta, A., Levine, S., & Finn, C. (2021) · arXiv (Cornell University) · Cited by 38
Multi-Ideology Multi-Class Extremism Classification Using Deep Learning Techniques
Gaikwad, M., Ahirrao, S. N., Kotecha, K. V., & Abraham, A. (2022) · IEEE Access · doi:10.1109/access.2022.3205744 · Cited by 36
Mirsadeghi, S. M. H., Bahşi, H., Vaarandi, R., & Inoubli, W. (2023) · IEEE Access · doi:10.1109/access.2023.3341755 · Cited by 42
Fayaz, S., Shah, S. Z. A., Din, N. M. U., Gul, N., & Assad, A. (2024) · Recent Advances in Computer Science and Communications · doi:10.2174/0126662558286875231215054324 · Cited by 40
Extending technological trajectories to latest technological changes by overcoming time lags
Hwang, S., & Shin, J. (2019) · Technological Forecasting and Social Change · doi:10.1016/j.techfore.2019.04.013 · Cited by 38
Zanjani, F. G., Zinger, S., Piepers, B., Mahmoudpour, S., Schelkens, P., & de With, P. H. N. (2019) · Journal of medical imaging · doi:10.1117/1.jmi.6.2.027501 · Cited by 38
“30 Here the negative and positive patches refer to the patches that have been labeled as normal and tumor patches, respectively.”
BackgroundImproving the Performance of VGG Through Different Granularity Feature Combinations
Zhou, Y., Chang, H., Lu, Y., Lu, X., & Zhou, R. (2020) · IEEE Access · doi:10.1109/access.2020.3031908 · Cited by 38
SMOTE-Based Weighted Deep Rotation Forest for the Imbalanced Hyperspectral Data Classification
Quan, Y., Zhong, X., Feng, W., Chan, J. C.-W., Li, Q., & Xing, M. (2021) · Remote Sensing · doi:10.3390/rs13030464 · Cited by 36
“However, most deep learning approaches, such as the convolutional neural network (CNN), have no algorithmic strategy for dealing with imbalanced data [63,65,66].”
BackgroundMachine Learning Approaches to Predict Chronic Lower Back Pain in People Aged over 50 Years
Shim, J., Ryu, K., Cho, E.-A., Ahn, J. H., Kim, H. K., Lee, Y.-J., & Lee, S. H. (2021) · Medicina · doi:10.3390/medicina57111230 · Cited by 36
“The AUROC from each machine learning model was plotted using the test dataset as a strong indicator of performance for classifiers in imbalanced datasets [21,22].”
MethodBenchmarking Deep Learning Models for Tooth Structure Segmentation
Schneider, L., Arsiwala-Scheppach, L. T., Krois, J., Meyer‐Lueckel, H., Bressem, K. K., Niehues, S. M., & Schwendicke, F. (2022) · Journal of Dental Research · doi:10.1177/00220345221100169 · Cited by 40
“Hence, it could be recommended to address this problem with weighted loss functions (Guerrero-Penã et al. 2018) or oversampling (Buda et al. 2018).”
BackgroundXu, Y., Yu, Z., & Chen, C. L. P. (2022) · IEEE Transactions on Neural Networks and Learning Systems · doi:10.1109/tnnls.2022.3177695 · Cited by 37
“Buda et al. [77] investigate the classification performance of combining CNN and various resampling methods in class imbalanced data, indicating that oversampling method is the most appropriate.”
MethodPosterior Re-calibration for Imbalanced DatasetsPreprint
Tian, J., Liu, Y., Glaser, N., Hsu, Y.-C., & Kira, Z. (2020) · arXiv (Cornell University) · doi:10.48550/arxiv.2010.11820 · Cited by 39
“2 Imbalanced CIFAR Classification We compare our methods with the state-of-the-art methods [1] [2] [9] [10] on CIFAR-10 and CIFAR100 dataset with two different imbalance types: long-tailed imbalance [1] and step imbalance [12] with two different imbalance ratios for each imbalance type.”
MethodMatsuzaka, Y., Hosaka, T., Ogaito, A., Yoshinari, K., & Uesawa, Y. (2020) · Molecules · doi:10.3390/molecules25061317 · Cited by 35
“is detrimental, and thresholding that compensates for prior class probabilities and oversampling to completely eliminate the imbalance dataset could be applied [50].”
BackgroundInfluentialA-CRNN-Based Method for Coherent DOA Estimation with Unknown Source Number
Yao, Y., Hong, L., & He, W. (2020) · Sensors · doi:10.3390/s20082296 · Cited by 35
“For the training dataset to a spatial filter, only one-sixth of the samples belong to the positive class while others are negative for L = 6, which can cause the class imbalance [27].”
BackgroundYang, G., Chen, G., Li, C., Fu, J., Guo, Y., & Hua, L. (2021) · Frontiers in Plant Science · doi:10.3389/fpls.2021.671134 · Cited by 37
“In general, category rebalancing strategies can be divided into two groups: re-sampling (Lee et al., 2016; Shen et al., 2016; Buda et al., 2018; Pouyanfar et al., 2018) and re-weighting (Huang et al., 2016, 2020; Wang et al., 2017; Cao et al., 2019; Cui et al., 2019).”
BackgroundMethodSumba, J. P. C., Alonso-Valerdi, L. M., & Ibarra-Zárate, D. I. (2022) · Applied Sciences · doi:10.3390/app12042158 · Cited by 40
Stasica, P., Nguyen, H. H., Granja, C. D. S., Kopeć, R., Marek, L., Oancea, C., Raczyński, Ł., Ruciński, A., Rydygier, M., Schubert, K. E., et al. (2023) · Physics in Medicine and Biology · doi:10.1088/1361-6560/acc9f8 · Cited by 35
“The limitations of AI in this work include the class imbalance problem which occurs when the distribution of classes in the training data set is unequally distributed and negatively affects the CNN ( Buda et al 2018 ) .”
BackgroundNo Matter How You Slice It: Machine Unlearning with SISA Comes at the Expense of Minority Classes
Koch, K., & Soll, M. (2023) · IEEE Conference on Secure and Trustworthy Machine Learning (SaTML) · doi:10.1109/satml54575.2023.00047 · Cited by 34
“In fact, even a 1 IS RUS baseline, which is S times faster than regular SISA, compared favorably to the SISA minority class error rates when we evaluated it.”
BackgroundInfluentialThe Equalization Losses: Gradient-Driven Training for Long-tailed Object Recognition
Tan, J., Li, B., Lu, X., Yao, Y., Yu, F., He, T., & Ouyang, W. (2023) · IEEE Transactions on Pattern Analysis and Machine Intelligence · doi:10.1109/tpami.2023.3298433 · Cited by 37
Time Series Forest Fire Prediction Based on Improved Transformer
Miao, X., Li, J., Mu, Y., He, C., Ma, Y., Chen, J., Wei, W., & Gao, D. (2023) · Forests · doi:10.3390/f14081596 · Cited by 36
MDCS: More Diverse Experts with Consistency Self-distillation for Long-tailed Recognition
Zhao, Q., Jiang, C., Hu, W., Zhang, F., & Liu, J. (2023) · IEEE/CVF International Conference on Computer Vision (ICCV) · doi:10.1109/iccv51070.2023.01065 · Cited by 35
“Re-sampling by simply over-sampling minority classes [3, 4, 38] and under-sampling by abandoning data for dominant classes [21, 16, 3].”
BackgroundApplicable and generalizable machine learning for intelligent welding in automotive manufacturing
Wang, P. E., Ghassemi-Armaki, H., Pour, M., Zhao, X., Ma, J., Sattari, K., & Carlson, B. E. (2025) · Welding in the World · doi:10.1007/s40194-025-01951-5 · Cited by 40
“ML models, determined by their data-driven learning nature, have a strong dependence on training data and are prone to low fidelity if training data have unbalanced or uncomprehensive coverage of application scenarios [26, 27].”
A comprehensive review on data-level methods for imbalanced data classification
Nikpour, B., Rahmati, F., Mirzaei, B., & Nezamabadi–pour, H. (2025) · Expert Systems with Applications · doi:10.1016/j.eswa.2025.128920 · Cited by 35
Medić, G., Kließ, M. K., Atallah, L., Weichert, J., Panda, S., Postma, M. J., & EL-Kerdi, A. (2019) · F1000Research · doi:10.12688/f1000research.20498.2 · Cited by 39
“Pham 2014 Retrospective case series single center France CHU de Caen, Caen & Hôpital Européen Georges-Pompidou, Paris NR (100) individuals suspected of having Venous thromboembolism NR”
BackgroundBoosting Minority Class Prediction on Imbalanced Point Cloud Data
Lin, H.-I., & Nguyen, M. C. (2020) · Applied Sciences · doi:10.3390/app10030973 · Cited by 34
“5 decision trees by Drummond [42] and other research [43,44] indicated that undersampling had a more pronounced effect than oversampling.”
BackgroundA Neural Network for Image Anomaly Detection with Deep Pyramidal Representations and Dynamic Routing
Mishra, P. K., Piciarelli, C., & Foresti, G. L. (2020) · International Journal of Neural Systems · doi:10.1142/s0129065720500604 · Cited by 33
Cost-Sensitive Regularization for Diabetic Retinopathy Grading from Eye Fundus Images
Galdrán, A., Dolz, J., Chakor, H., Lombaert, H., & Ayed, I. B. (2020) · Lecture notes in computer science · doi:10.1007/978-3-030-59722-1_64 · Cited by 34
“formance stagnates in the validation set, and the training is stopped after 10 epochs of no further improvement. In addition, to mitigate the impact of class imbalance, we oversample minority classes [19]. 3 Experimental Validation In this section we describe the experimental setting we follow in order to validate our…”
MethodInfluentialHow to evaluate deep learning for cancer diagnostics – factors and recommendations
Daneshjou, R., He, B. D., Ouyang, D., & Zou, J. (2021) · Biochimica et Biophysica Acta (BBA) - Reviews on Cancer · doi:10.1016/j.bbcan.2021.188515 · Cited by 35
Yang, D., Li, T., Liu, M.-T., Li, X., & Chen, B. (2021) · Ecological Informatics · doi:10.1016/j.ecoinf.2021.101350 · Cited by 37
Seizure Prediction using Convolutional Neural Networks and Sequence Transformer Networks
Chen, R., & Parhi, K. K. (2021) · Annual International Conference of the IEEE Engineering in Medicine and Biology Society (EMBC) · doi:10.1109/embc46164.2021.9629732 · Cited by 34
“This is undesirable as highly imbalanced datasets often lead to poor performance in classification with a CNN [11].”
BackgroundDeep Transfer Learning with Enhanced Feature Fusion for Detection of Abnormalities in X-ray Images
Alammar, Z., Alzubaidi, L., Zhang, J., Li, Y., Lafta, W. M., & Gu, Y. (2023) · Cancers · doi:10.3390/cancers15154007 · Cited by 40
“However, unbalanced data remain a limitation in certain studies [18].”
BackgroundApplication of Decision Rules for Handling Class Imbalance in Semantic SegmentationPreprint
Chan, R., Rottmann, M., Hüger, F., Schlicht, P., & Gottschalk, H. (2019) · arXiv (Cornell University) · doi:10.48550/arxiv.1901.08394 · Cited by 34
“Oversampling and undersampling strategies have been proposed in [4, 14, 20].”
BackgroundRethinking of Pedestrian Attribute Recognition: Realistic Datasets with Efficient MethodPreprint
Jia, J., Huang, H., Yang, W., Chen, X., & Huang, K. (2020) · arXiv (Cornell University) · doi:10.48550/arxiv.2005.11909 · Cited by 45
Adding Stochastic Negative Examples into Machine Learning Improves Molecular Bioactivity Prediction
Cáceres, E. L., Mew, N. C., & Keiser, M. J. (2020) · Journal of Chemical Information and Modeling · doi:10.1021/acs.jcim.0c00565 · Cited by 33
Learning From Imbalanced Data With Deep Density Hybrid Sampling
Liu, C., & Chang, Y. (2022) · IEEE Transactions on Systems Man and Cybernetics Systems · doi:10.1109/tsmc.2022.3151394 · Cited by 34
“Buda et al. [25] have conducted extensive experiments on three benchmark datasets to investigate the effects of imbalance on convolutional neural networks (CNNs).”
BackgroundRauch, L., & Braml, T. (2023) · Applied Sciences · doi:10.3390/app13169146 · Cited by 36
“A compilation of cost-sensitive learning approaches for neural networks is given by Buda et al. [103].”
BackgroundMethodInfluentialVariational Continuous Label Distribution Learning for Multi-Label Text Classification
Zhao, X., An, Y., Xu, N., & Geng, X. (2023) · IEEE Transactions on Knowledge and Data Engineering · doi:10.1109/tkde.2023.3323401 · Cited by 35
“It is based on undersampling [46], [47], [48] or over-sampling [48], [49], [50], which removes samples associated with the majority label or generates new samples associated with the minority label.”
BackgroundDemirci, M. Y., Beșlı, N., & Gümüşçü, A. (2024) · Expert Systems with Applications · doi:10.1016/j.eswa.2024.124230 · Cited by 33
Li, Y., Yang, Y., Song, P., Duan, L., & Ren, R. (2025) · Scientific Reports · doi:10.1038/s41598-025-09506-w · Cited by 33
“The main methods include cost-sensitive learning 21,22 , threshold moving strategies 23,24 , ensemble learning 25,26 and neural networks 27,28 .”
Mining and Classifying Aviation Accident Reports
Srinivasan, P., Nagarajan, V., & Mahadevan, S. (2019) · AIAA AVIATION Forum · doi:10.2514/6.2019-2938 · Cited by 34
Pulsar candidate classification with deep convolutional neural networks
Wang, Y.-C., Li, M.-T., Pan, Z.-C., & Zheng, J.-H. (2019) · Research in Astronomy and Astrophysics · doi:10.1088/1674-4527/19/9/133 · Cited by 34
Real-Time Driving Scene Semantic Segmentation
Wang, W., Fu, Y., Pan, Z., Li, X., & Zhuang, Y. (2020) · IEEE Access · doi:10.1109/access.2020.2975640 · Cited by 33
Alani, A. A., Cosma, G., & Taherkhani, A. (2020) · International Joint Conference on Neural Networks (IJCNN) · doi:10.1109/ijcnn48605.2020.9207697 · Cited by 33
“Methods of dealing with the class imbalance problem are well studied for classical ML models and these methods can be divided into two main approaches [6].”
BackgroundMethodInfluentialToward hierarchical classification of imbalanced data using random resampling algorithms
Pereira, R. M., Costa, Y. M. G., & Junior, C. N. S. (2021) · Information Sciences · doi:10.1016/j.ins.2021.07.033 · Cited by 33
GRA-GAN: Generative adversarial network for image style transfer of Gender, Race, and age
Kim, Y. H., Nam, S. H., Hong, S. B., & Park, K. R. (2022) · Expert Systems with Applications · doi:10.1016/j.eswa.2022.116792 · Cited by 34
Bodendorf, F., Sauter, M., & Franke, J. E. (2022) · International Journal of Production Economics · doi:10.1016/j.ijpe.2022.108708 · Cited by 40
Norman, D. L., Bischoff, P. H., Wearn, O. R., Ewers, R. M., Rowcliffe, J. M., Evans, B. C., Sethi, S. S., Chapman, P. M., & Freeman, R. (2022) · Methods in Ecology and Evolution · doi:10.1111/2041-210x.14031 · Cited by 36
“Class imbalance is common within camera trap datasets, and has been shown to impair the performance of neural networks (Buda et al., 2018).”
BackgroundMultimodal Analysis of Unbalanced Dermatological Data for Skin Cancer Recognition
Lyakhov, P. A., Lyakhova, U. A., & Ивановна, К. Д. (2023) · IEEE Access · doi:10.1109/access.2023.3336289 · Cited by 36
Addressing imbalanced data for machine learning based mineral prospectivity mapping
Farahnakian, F., Sheikh, J., Zelioli, L., Nidhi, D. K., Seppä, I., Ilo, R., Nevalainen, P., & Heikkonen, J. (2024) · Ore Geology Reviews · doi:10.1016/j.oregeorev.2024.106270 · Cited by 32
Ye, W., Lin, Y., Li, M., Liu, Q., & Pan, D. Z. (2019) · Asia and South Pacific Design Automation Conference (ASP-DAC) · doi:10.1145/3287624.3288746 · Cited by 32
“It simply replicates randomly selected samples from minority classes, but this approach can increase the time necessary to build a classifier, and may even lead to overfitting [27].”
BackgroundInfluence of image quality on the identification of psyllids using convolutional neural networks
Barbedo, J. G. A., & Castro, G. B. (2019) · Biosystems Engineering · doi:10.1016/j.biosystemseng.2019.04.007 · Cited by 32
Fine-grained flow classification using deep learning for software defined data center networks
Liu, W., Cai, J., Wang, Y., Chen, Q. C., & Zeng, J. (2020) · Journal of Network and Computer Applications · doi:10.1016/j.jnca.2020.102766 · Cited by 32
1D Convolutional Neural Networks for Detecting Nystagmus
Newman, J. L., Phillips, J. S., & Cox, S. J. (2020) · IEEE Journal of Biomedical and Health Informatics · doi:10.1109/jbhi.2020.3025381 · Cited by 32
“Experimentally, oversampling has been shown to outperform undersampling [25], [26], especially when applied to large class imbalances and when training neural networks.”
BackgroundFrom generalized zero-shot learning to long-tail with class descriptors
Samuel, D., Atzmon, Y., & Chechik, G. (2021) · IEEE Winter Conference on Applications of Computer Vision · doi:10.1109/wacv48630.2021.00033 · Cited by 34
“Long-tail learning: Learning with unbalanced data causes models to favor head classes [6].”
BackgroundKim, T.-S., Kim, J., Choi, H. S., Kim, E. S., Keum, B., Jeen, Y. T., Lee, H. S., Chun, H. J., Han, S. Y., Kim, D. U., et al. (2021) · Scientific Reports · doi:10.1038/s41598-021-87737-3 · Cited by 35
Im, S., Hyeon, J., Rha, E. Y., Lee, J., Choi, H., Jung, Y., & Kim, T. J. (2021) · Sensors · doi:10.3390/s21103500 · Cited by 33
“Oversampling images to adjust the ratio of each images should especially be performed in the small dataset [52].”
BackgroundA Class-Imbalanced Deep Learning Fall Detection Algorithm Using Wearable Sensors
Zhang, J., Li, J., & Wang, W. (2021) · Sensors · doi:10.3390/s21196511 · Cited by 32
“Moreover, Buda et al. [31] experimented with adjusting the CNN output threshold on a dataset preprocessed by the SMOTE algorithm.”
Label-Aware Distribution Calibration for Long-Tailed Classification
Wang, C., Gao, S., Wang, P., Gao, C., Pei, W., Pan, L., & Xu, Z. (2022) · IEEE Transactions on Neural Networks and Learning Systems · doi:10.1109/tnnls.2022.3213522 · Cited by 34
“Resampling incorporates two approach types, i.e., over-sampling tail classes [44], [45], [46] and under-sampling head classes [45], [47].”
MethodMS-ALN: Multiscale Attention Learning Network for Pest Recognition
Feng, F., Dong, H., Zhang, Y., Zhang, Y., & Li, B. (2022) · IEEE Access · doi:10.1109/access.2022.3167397 · Cited by 34
Barriga, J. A., Blanco-Cipollone, F., Trigo-Córdoba, E., García-Tejero, I., & Clemente, P. J. (2022) · Expert Systems with Applications · doi:10.1016/j.eswa.2022.118255 · Cited by 35
A Hybrid Sampling Approach for Imbalanced Binary and Multi-Class Data Using Clustering Analysis
Palli, A. S., Jaafar, J., Hashmani, M. A., Gomes, H. M., & Gilal, A. R. (2022) · IEEE Access · doi:10.1109/access.2022.3218463 · Cited by 33
Yan, M., Hui, S. C., & Li, N. (2023) · Information Sciences · doi:10.1016/j.ins.2023.01.074 · Cited by 33
Wang, X., Gao, Y., Hou, J., Yang, J., Smits, K. M., & He, H. (2024) · Journal of Hydrology · doi:10.1016/j.jhydrol.2024.130950 · Cited by 32
A Comprehensive Review and Assessment of Cybersecurity Vulnerability Detection Methodologies
Bennouk, K., Aali, N. A., Idrissi, Y. E. B. E., Sebai, B., Faroukhi, A. Z., & Mahouachi, D. (2024) · Journal of Cybersecurity and Privacy · doi:10.3390/jcp4040040 · Cited by 48
Shirvan, M. H., Moattar, M. H., & Hosseinzadeh, M. (2025) · Applied Soft Computing · doi:10.1016/j.asoc.2024.112677 · Cited by 31
Impact of imbalanced features on large datasets
Albattah, W., & Khan, R. U. (2025) · Frontiers in Big Data · doi:10.3389/fdata.2025.1455442 · Cited by 45
“The CB approach adjusts the weights of instances in the dataset so that each class contributes equally to the total weight (Buda et al., 2018).”
Wang, J., Zhang, W., & Zhou, J. (2020) · Industrial & Engineering Chemistry Research · doi:10.1021/acs.iecr.9b06298 · Cited by 28
Jones, E., Wong, S., Milton, A., Sclauzero, J., Whittenbury, H., & McDonnell, M. D. (2020) · Remote Sensing · doi:10.3390/rs12060934 · Cited by 32
Impact of teachers' grading policy on the identification of at-risk students in learning analytics
Lu, O. H. T., Huang, A. Y. Q., & Yang, S. J. H. (2021) · Comput. Educ. · doi:10.1016/j.compedu.2020.104109 · Cited by 29
Multiscale CNN with compound fusions for false positive reduction in lung nodule detection
Mittapalli, P. S., & Thanikaiselvan, V. (2021) · Artificial Intelligence in Medicine · doi:10.1016/j.artmed.2021.102017 · Cited by 31
Tae, K. H., & Whang, S. E. (2021) · ACM SIGMOD International Conference on Management of Data (SIGMOD) · doi:10.1145/3448016.3452792 · Cited by 35
“We use imbalance ratio [11], which is the maximum ratio between any two slices in S , as a proxy for bias.”
MethodAlkhodari, M., Rashid, M., Mukit, M. A., Ahmed, K. I. U., Mostafa, R., Parveen, S., & Khandoker, A. H. (2021) · IEEE Access · doi:10.1109/access.2021.3107687 · Cited by 35
Key Point Sensitive Loss for Long-tailed Visual Recognition
Li, M., Cheung, Y., & Hu, Z. (2022) · IEEE Transactions on Pattern Analysis and Machine Intelligence · doi:10.1109/tpami.2022.3196044 · Cited by 32
“The most important two manners of re-sampling meth-ods include under-sampling the head classes [9], [11], [12] and over-sampling the tail classes [6], [7], [8].”
BackgroundJin, W., Fatehi, M., Guo, R., & Hamarneh, G. (2024) · Artificial Intelligence in Medicine · doi:10.1016/j.artmed.2023.102751 · Cited by 33
A Deep Learning Approach to Infer Employment Status of Passengers by Using Smart Card Data
Zhang, Y., & Cheng, T. (2019) · IEEE Transactions on Intelligent Transportation Systems · doi:10.1109/tits.2019.2896460 · Cited by 31
Xia, W., Ma, C., Liu, J., Liu, S., Fu, C., Yang, Z. C., & Duan, J. (2019) · Remote Sensing · doi:10.3390/rs11212523 · Cited by 34
“In a large number of analyzed scenarios, under sampling showed decrease in performance as compared to the original dataset [39].”
BackgroundInfluentialTowards Calibrated Model for Long-Tailed Visual Recognition from Prior PerspectivePreprint
Xu, Z., Chai, Z., & Yuan, C. (2021) · arXiv (Cornell University) · doi:10.48550/arxiv.2111.03874 · Cited by 31
“Intuitively, under-sampling the head [20, 6, 22] or over-sampling the tail [11, 6, 48, 54, 7] can improve the inconsistent performance of imbalanced datasets but tend to either weaken the head or over-fitting the tail.”
Deeply Supervised Skin Lesions Diagnosis With Stage and Branch Attention
Dai, W., Liu, R., Wu, T., Wang, M., Yin, J., & Liu, J. (2023) · IEEE Journal of Biomedical and Health Informatics · doi:10.1109/jbhi.2023.3308697 · Cited by 34
“Such relatively high imbalance ratios can remarkably reduce training performance, according to previous research [50].”
ResultEnhanced Long-Tailed Recognition With Contrastive CutMix Augmentation
Pan, H., Guo, Y. F., Yu, M., & Chen, J. (2024) · IEEE Transactions on Image Processing · doi:10.1109/tip.2024.3425148 · Cited by 37
“Over-sampling [14], [15] is one of the most common methods to alleviate data scarcity, it emphasizes the tail classes and increases the instance number of the tail classes to reduce the imbalance.”
MethodAbuAlkebash, H., Saleh, R. A. A., & Ertunç, H. M. (2025) · Biomedical Signal Processing and Control · doi:10.1016/j.bspc.2025.107934 · Cited by 31
Saleh, K., Hossny, M., & Nahavandi, S. (2019) · Neurocomputing · doi:10.1016/j.neucom.2019.12.091 · Cited by 30
Ultrasonic signal denoising based on autoencoder
Gao, F., Li, B., Chen, L., Wei, X., Shang, Z., & He, C. (2020) · Review of Scientific Instruments · doi:10.1063/1.5136269 · Cited by 29
Avanzo, M., Pirrone, G., Vinante, L., Caroli, A., Stancanello, J., Drigo, A., Massarut, S., Mileto, M., Urbani, M., Trovò, M., et al. (2020) · Frontiers in Oncology · doi:10.3389/fonc.2020.00490 · Cited by 30
“The effect of imbalanced datasets on ML performance is detrimental (21, 22), and there are two methods for overcoming this issue, namely under-sampling and over-sampling, of which the latter has been proven to be more effective in ML (21).”
MethodHybrid SVM-CNN Classification Technique for Human–Vehicle Targets in an Automotive LFMCW Radar
Wu, Q., Gao, T. F., Lai, Z., & Li, D. (2020) · Sensors · doi:10.3390/s20123504 · Cited by 32
“Reference [23] showed the network less of what it has already learned and more of what it does not understand yet by over-sampling the low performing classes and under-sampling the high performing classes.”
BackgroundMethodChallenges for ocular disease identification in the era of artificial intelligence
Gour, N., Tanveer, M., & Khanna, P. (2022) · Neural Computing and Applications · doi:10.1007/s00521-021-06770-5 · Cited by 30
“The majority class is under-sampled or the minority class is oversampled to create a well-balanced training set [147].”
BackgroundShaaf, Z. F., Jamil, M. M. A., Ambar, R., Alattab, A. A., Yahya, A. A., & Asiri, Y. (2022) · Diagnostics · doi:10.3390/diagnostics12020414 · Cited by 30
“[45] provided a thorough analysis of the CNN class imbalance problem.”
A hybrid imbalanced classification model based on data density
Shi, S., Li, J., Zhu, D., Yang, F. O., & Xu, Y. (2022) · Information Sciences · doi:10.1016/j.ins.2022.12.046 · Cited by 36
Learning Long-Range Perception Using Self-Supervision From Short-Range Sensors and Odometry
Nava, M., Guzzi, J., Chavéz-García, R. O., Gambardella, L. M., & Giusti, A. (2019) · IEEE Robotics and Automation Letters · doi:10.1109/lra.2019.2894849 · Cited by 32
“vor of negative labels; a potential countermeasure, which is not necessary in our case, is to implement a cost-sensitive loss [22].”
Background
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
