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 601–700 of 1,001
Secure UAV‐Based System to Detect Small Boats Using Neural Networks
Lodeiro-Santiago, M., Caballero‐Gil, P., Aguasca-Colomo, R., & Caballero‐Gil, C. (2019) · Complexity · doi:10.1155/2019/7206096 · Cited by 29
A comparative study and analysis of developer communications on Slack and Gitter
Parra, E., Alahmadi, M. D., Ellis, A., & Haiduc, S. (2022) · Empirical Software Engineering · doi:10.1007/s10664-021-10095-1 · Cited by 29
“To mitigate the impact of class imbalance, we applied robust over-sampling approaches as well as under-sampling approaches, since it has been shown that over-sampling can cause over-fitting for classical machine learning models, especially for minority classes (Buda et al. 2018).”
MethodIntelligent Health Risk and Disease Prediction Using Optimized Naive Bayes Classifier
Alamer, L. (2023) · Journal of Internet Services and Information Security · doi:10.58346/jisis.2023.i1.001 · Cited by 29
“For each subset values classified based on clinical, administrative [5], academic [6], educational [7] and disease background [8].”
BackgroundClassification of Liver Fibrosis From Heterogeneous Ultrasound Image
Joo, Y., Park, H.-C., Lee, O., Yoon, C., Choi, M. H., & Choi, C. (2023) · IEEE Access · doi:10.1109/access.2023.3240216 · Cited by 29
“Data augmentation was performed to mitigate the class imbalance problem of the five-level classification [25].”
MethodUncertainty-Guided Boundary Learning for Imbalanced Social Event Detection
Ren, J., Peng, H. M., Jiang, L., Liu, Z., Wu, J. X., Yu, Z., & Yu, P. S. (2023) · IEEE Transactions on Knowledge and Data Engineering · doi:10.1109/tkde.2023.3324510 · Cited by 29
“Two common strategies are: 1) data re-sampling [4], [5], [6], whose core idea is to increase the numbers of samples in the tail classes or decrease the samples of those head classes; and 2) loss reweighting [7], [8], [9], [10], in which weights assigning to tail classes are larger.”
BackgroundMethodFeature augmentation for imbalanced classification with conditional mixture WGANs
Zhang, Y., Sun, B., Xiao, Y., Xiao, R., & Wei, Y. (2019) · Signal Processing Image Communication · doi:10.1016/j.image.2019.03.010 · Cited by 28
Liu, Z., Wu, B.-H., Li, D., Xiao, P., & Mao, J.-W. (2020) · Sensors · doi:10.3390/s20082297 · Cited by 29
“Oversampling can reduce the degree of data imbalance at the data level by constructing new samples, but artificially synthesized minority samples may increase the risk of overlearning in minority samples [21].”
BackgroundA three-level classification of French tweets in ecological crises
Kozlowski, D., Lannelongue, E., Saudemont, F., Benamara, F., Mari, A., Moriceau, V., & Boumadane, A. (2020) · Information Processing & Management · doi:10.1016/j.ipm.2020.102284 · Cited by 34
“Another interesting perspective is to explore the class imbalance problem in the context of deep learning (Buda et al., 2017, Johnson and Khoshgoftaar, 2019, Koziarski et al., 2019).”
BackgroundDiscriminative feature generation for classification of imbalanced data
Suh, S., Lukowicz, P., & Lee, Y. O. (2021) · Pattern Recognition · doi:10.1016/j.patcog.2021.108302 · Cited by 29
“To evaluate the proposed method under the class-imbalanced condition, we adopt the step imbalance type described in [3].”
BackgroundMethodBoosting-GNN: Boosting Algorithm for Graph Networks on Imbalanced Node Classification
Shi, S., Qiao, K., Yang, S., Wang, L., Chen, J., & Yan, B. (2021) · Frontiers in Neurorobotics · doi:10.3389/fnbot.2021.775688 · Cited by 33
“Resampling Resampling can be specifically divided into two types: 1) Oversampling by copying data in minority classes (Buda et al., 2018; Byrd and Lipton, 2019).”
BackgroundWater Quality Assessment Tool for On-Site Water Quality Monitoring
Olatinwo, S. O., & Joubert, T.-H. (2024) · IEEE Sensors Journal · doi:10.1109/jsen.2024.3383887 · Cited by 31
“To address this issue, we employed the SMOTE augmentation method [30], [31] from the imbalanced-learn library in Python.”
MethodA Deep Learning Framework for Tactile Recognition of Known as Well as Novel Objects
Abderrahmane, Z., Ganesh, G., Crosnier, A., & Cherubini, A. (2019) · IEEE Transactions on Industrial Informatics · doi:10.1109/tii.2019.2898264 · Cited by 29
“This decreases the classification accuracy due to the imbalance in the training set [13], [14].”
BackgroundZhao, Y., Ren, H., Zhang, Y., Wang, C., & Long, Y. (2024) · Optics & Laser Technology · doi:10.1016/j.optlastec.2024.110648 · Cited by 30
LTGC: Long-Tail Recognition via Leveraging LLMs-Driven Generated Content
Zhao, Q., Dai, Y., Li, H., Hu, W., Zhang, F., & Liu, 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/cvpr52733.2024.01845 · Cited by 35
“To address these challenges, numerous approaches have emerged within the field of long-tailed recognition, such as re-sampling [5, 19], loss re-weighting [1, 6, 12, 23, 26, 27, 29, 39, 48], ensemble learning [42, 47, 56], decoupling [20], and contrastive learning [11, 25, 40, 59].”
Cattle Detection Using Oblique UAV Images
Barbedo, J. G. A., Koenigkan, L. V., & Santos, P. M. (2020) · Drones · doi:10.3390/drones4040075 · Cited by 30
Handling Imbalanced Dataset Classification in Machine Learning
Yadav, S., & Bhole, G. P. (2020) · IEEE Pune Section International Conference (PuneCon) · doi:10.1109/punecon50868.2020.9362471 · Cited by 28
Weinzierl, S., Wolf, V., Pauli, T., Beverungen, D., & Matzner, M. (2021) · Journal of Business Analytics · doi:10.1080/2573234x.2021.1978337 · Cited by 35
“In particular, the learning process of a CNN, as used here, can be negatively affected by class imbalance (Buda et al., 2018).”
MethodGeometric imbalanced deep learning with feature scaling and boundary sample mining
Wang, Z., Dong, Q., Guo, W., Li, D., Zhang, J., & Du, W. (2022) · Pattern Recognition · doi:10.1016/j.patcog.2022.108564 · Cited by 27
Ma, Y., Liu, M., Zhang, Y., Zhang, B., Xu, K., Zou, B., & Huang, Z. (2022) · Remote Sensing · doi:10.3390/rs14164103 · Cited by 28
Data augmentation based on multiple oversampling fusion for medical image segmentation
Wu, L., Zhuang, J., Chen, W., Tang, Y., Hou, C., Li, C., Zhong, Z., & Luo, S. (2022) · PLoS ONE · doi:10.1371/journal.pone.0274522 · Cited by 28
“[16] proved that oversampling to alleviate category imbalances is a suitable technique for deep learning.”
BackgroundConstructing Balance from Imbalance for Long-Tailed Image Recognition
Xu, Y., Li, Y., Li, J., & Lu, C. (2022) · Lecture notes in computer science · doi:10.1007/978-3-031-20044-1_3 · Cited by 31
Neural collapse inspired attraction–repulsion-balanced loss for imbalanced learning
Xie, L., Yang, Y., Cai, D., & He, X. (2023) · Neurocomputing · doi:10.1016/j.neucom.2023.01.023 · Cited by 37
“Early works [1,14,15,2] re-sample the data (oversample the samples of minor classes or under-sample the samples of major classes or both) to balance the data distribution.”
BackgroundA Transfer Learning and Optimized CNN Based Maritime Vessel Classification System
Salem, M. H., Li, Y., Liu, Z., & AbdelTawab, A. M. (2023) · Applied Sciences · doi:10.3390/app13031912 · Cited by 28
Long-Tailed Food Classification
He, J., Lin, L., Eicher‐Miller, H. A., & Zhu, F. M. (2023) · Nutrients · doi:10.3390/nu15122751 · Cited by 28
“In this section, we first evaluate our head class undersampling by comparing RUS [23] with (i) replacing with Herding selection (HUS) and (ii) applying knowledge distillation Datasets Food101-LT VFN-LT”
BackgroundMethodInfluentialEltehewy, R., Abouelfarag, A. A., & Saleh, S. N. (2023) · ISPRS International Journal of Geo-Information · doi:10.3390/ijgi12060245 · Cited by 28
“The ratio (ρ) can be expressed as follows, where Ci is a set of examples f r ach class i [37].”
BackgroundTwo-stage fine-tuning with ChatGPT data augmentation for learning class-imbalanced data
ValizadehAslani, T., Shi, Y., Wang, J., Ren, P., Zhang, Y., Hu, M., Zhao, L., & Liang, H. (2024) · Neurocomputing · doi:10.1016/j.neucom.2024.127801 · Cited by 43
Genomics‐based plant disease resistance prediction using machine learning
Upadhyaya, S. R., Danilevicz, M. F., Dolatabadian, A., Neik, T. X., Zhang, F., Al-Mamun, H. A., Bennamoun, M., Batley, J., & Edwards, D. (2024) · Plant Pathology · doi:10.1111/ppa.13988 · Cited by 36
Tufail, R., Tassinari, P., & Torreggiani, D. (2025) · Remote Sensing · doi:10.3390/rs17183207 · Cited by 34
Buda, M., AlBadawy, E. A., Saha, A., & Mazurowski, M. A. (2020) · Radiology Artificial Intelligence · doi:10.1148/ryai.2019180050 · Cited by 28
“To alleviate the problem of imbalance, we applied random minority oversampling to make class distribution uniform (12).”
MethodSelf-citationImpact of Base Dataset Design on Few-Shot Image Classification
Sbai, O., Couprie, C., & Aubry, M. (2020) · Lecture notes in computer science · doi:10.1007/978-3-030-58517-4_35 · Cited by 27
“For instance, the study of [3] on class imbalance highlights over-sampling or under-sampling strategies that are privileged in many works.”
BackgroundConversion of adverse data corpus to shrewd output using sampling metrics
Ashraf, S., Saleem, S., Ahmed, T., Aslam, Z., & Muhammad, D. (2020) · Visual Computing for Industry Biomedicine and Art · doi:10.1186/s42492-020-00055-9 · Cited by 27
Parkinson’s Disease Detection Using Ensemble Architecture from MR Images
Mostafa, T. A., & Cheng, I. (2020) · IEEE International Conference on Bioinformatics and Bioengineering (BIBE) · doi:10.1109/bibe50027.2020.00167 · Cited by 27
“over classifying the majority class [14].”
Ni, M., Wang, H., Liu, X., Liao, Y., Fu, L., Wu, Q., Mu, J., Chen, X., & Li, J. (2021) · Sensors · doi:10.3390/s21020638 · Cited by 27
“• The droplet deposition rate of the same spray state is only recorded once to avoid the regression analysis bias caused by the data size [63].”
BackgroundSchultheiß, M., Schmette, P., Bodden, J. H., Aichele, J., Müller-Leisse, C., Gassert, F. G., Gassert, F. T., Gawlitza, J. F., Hofmann, F. C., Sasse, D., et al. (2021) · Scientific Reports · doi:10.1038/s41598-021-94750-z · Cited by 33
Procrustean Training for Imbalanced Deep Learning
Ye, H.-J., Zhan, D., & Chao, W. (2021) · IEEE/CVF International Conference on Computer Vision (ICCV) · doi:10.1109/iccv48922.2021.00016 · Cited by 30
“Deep neural networks [21, 25, 37, 55, 56] trained with class-imbalanced data, of which we have ample data for some “major” classes and limited data for the other “minor” classes, are known to perform poorly on minor classes [3, 45, 61].”
BackgroundMylonas, N., Malounas, I., Mouseti, S., Vali, E., Espejo-García, B., & Fountas, S. (2021) · Smart Agricultural Technology · doi:10.1016/j.atech.2021.100028 · Cited by 27
Lee, A. L. S., To, C. C. K., Lee, A. L.-H., Li, J. J. X., & Chan, R. (2022) · Informatics in Medicine Unlocked · doi:10.1016/j.imu.2022.100850 · Cited by 27
Gernat, T., Jagla, T., Jones, B. M., Middendorf, M., & Robinson, G. E. (2023) · Scientific Reports · doi:10.1038/s41598-022-26825-4 · Cited by 32
“Because CNNs learn to classify images better if the number of positive and negative examples is more balanced 32 , and manual annotation is labor intensive, additional images were selected by also requiring that bees needed to face each other, which is necessary for trophallaxis.”
BackgroundA novel graph oversampling framework for node classification in class-imbalanced graphs
Xia, R., Zhang, C., Zhang, Y., Liu, X., & Yang, B. X. (2024) · Science China Information Sciences · doi:10.1007/s11432-023-3897-2 · Cited by 28
Predicting future fallers in Parkinson’s disease using kinematic data over a period of 5 years
Sotirakis, C., Brzezicki, M. A., Patel, S. H., Conway, N., FitzGerald, J. J., & Antoniades, C. A. (2024) · npj Digital Medicine · doi:10.1038/s41746-024-01311-5 · Cited by 26
Convolutional Neural Network for Imbalanced Data Classification of Silicon Wafer Defects
Batool, U., Shapiai, M. I., Fauzi, H., & Fong, J. X. (2020) · IEEE International Colloquium on Signal Processing & Its Applications (CSPA) · doi:10.1109/cspa48992.2020.9068669 · Cited by 26
“A dataset is considered imbalanced if its class sizes are unequal [21], [22], [23].”
BackgroundAutee, P., Bagwe, S., Shah, V., & Srivastava, K. (2020) · Physical and Engineering Sciences in Medicine · doi:10.1007/s13246-020-00952-6 · Cited by 28
Shaw, D., Chen, H., Xie, M., & Jiang, T. (2021) · BMC Bioinformatics · doi:10.1186/s12859-020-03914-7 · Cited by 27
Swiecicki, A., Konz, N., Buda, M., & Mazurowski, M. A. (2021) · Scientific Reports · doi:10.1038/s41598-021-89626-1 · Cited by 28
“Such imbalance in the training dataset causes significant problems when training deep learning algorithms and has been shown to negatively affect model performance 4 .”
BackgroundSelf-citationA Deep convolutional neural network with residual blocks for wafer map defect pattern recognition
Wang, F., Chou, J., & Amogne, Z. E. (2021) · Quality and Reliability Engineering International · doi:10.1002/qre.2983 · Cited by 27
Learning Imbalanced Datasets With Maximum Margin Loss
Kang, H., Vu, T., & Yoo, C. D. (2021) · Proceedings - International Conference on Image Processing · doi:10.1109/icip42928.2021.9506389 · Cited by 27
“The two classical approaches for learning long-tailed data are re-weighting [6] the losses of the examples and re-sampling (over-sampling the minority classes [7] and under-sampling the majority classes [8]) the examples in the SGD mini-batch.”
BackgroundMethodMultiview Wasserstein generative adversarial network for imbalanced pearl classification
Gao, S. L., Dai, Y., Li, Y., Liu, K., Chen, K., & Liu, Y. (2022) · Measurement Science and Technology · doi:10.1088/1361-6501/ac6224 · Cited by 26
Deep semantic segmentation for visual scene understanding of soil types
Zamani, V., Taghaddos, H., Gholipour, Y., & Pourreza, H. R. (2022) · Automation in Construction · doi:10.1016/j.autcon.2022.104342 · Cited by 26
Kiessner, A.-K., Schirrmeister, R. T., Gemein, L. A. W., Boedecker, J., & Tonio Ball, A. (2023) · NeuroImage Clinical · doi:10.1016/j.nicl.2023.103482 · Cited by 25
A profitable trading algorithm for cryptocurrencies using a Neural Network model
Parente, M., Rizzuti, L., & Trerotola, M. (2023) · Expert Systems with Applications · doi:10.1016/j.eswa.2023.121806 · Cited by 26
Research on Deep Learning Model of Feature Extraction Based on Convolutional Neural Network
Liu, H., Li, I., Liang, Y., Sun, D., Yang, Y., & Yang, H. (2024) · IEEE International Conference on Image Processing and Computer Applications (ICIPCA) · doi:10.1109/icipca61593.2024.10709168 · Cited by 26
“The method of "teaching" by the teacher is proposed to realize the learning of various parameters required in the modeling process of the teacher [26].”
MethodSerendipity—A Machine-Learning Application for Mining Serendipitous Drug Usage From Social Media
Ru, B., Li, D., Hu, Y., & Yao, L. (2019) · IEEE Transactions on NanoBioscience · doi:10.1109/tnb.2019.2909094 · Cited by 26
“Moreover, neural networks are sensitive to imbalanced data [44].”
Customer Satisfaction Estimation in Contact Center Calls Based on a Hierarchical Multi-Task Model
Ando, A., Masumura, R., Kamiyama, H., Kobashikawa, S., Aono, Y., & Toda, T. (2020) · IEEE/ACM Transactions on Audio Speech and Language Processing · doi:10.1109/taslp.2020.2966857 · Cited by 26
Jang, M., Bae, H., Kim, M., Park, S. Y., Son, A. Y., Choi, S. J., Choe, J., Choi, H. Y., Hwang, H. J., Noh, H. N., et al. (2023) · Scientific Reports · doi:10.1038/s41598-023-28175-1 · Cited by 35
“The effects of class imbalance on the performance of CNNs were examined, where it was determined to decrease the performance of CNN 35,36 .”
BackgroundInfluentialA Data Augmentation Methodology to Reduce the Class Imbalance in Histopathology Images
Guerrero, R. E. D., Carvalho, L., Bocklitz, T. W., Popp, J., & Oliveira, J. L. (2024) · Journal of Imaging Informatics in Medicine · doi:10.1007/s10278-024-01018-9 · Cited by 30
“…primary approaches: (a) classifier-level solutions, e.g., cost-sensitive learning [9], thresholding methods [10], or one-class classification [11]; (b) data-level solutions, such as oversampling [12, 13], or undersampling [14]; (c) hybrid [15], a combination of the two previous approaches [16, 17].”
BackgroundSlevitch, L. (2024) · Journal of Hospitality & Tourism Research · doi:10.1177/10963480241230957 · Cited by 31
Aghamohammadesmaeilketabforoosh, K., Nikan, S., Antonini, G., & Pearce, J. M. (2024) · Foods · doi:10.3390/foods13121869 · Cited by 25
“To address the imbalanced class distribution in the dataset, each class was assigned weights using PyTorch’s Weighted Random Sampler [14] to ensure nearly equal impact during the training process.”
BackgroundMethodMing, R., Abdelrahman, O., Innab, N., & Hanafy, M. (2024) · PeerJ Computer Science · doi:10.7717/peerj-cs.2088 · Cited by 25
“To address this issue, we employed oversampling technique ( Buda, Maki & Mazurowski, 2018 ), which involved randomly replicating instances from the minority category to match the number of instances in the majority category, as demonstrated in Table 1.”
MethodSun, J., Yan, S., Yao, X., Gao, B., & Yang, J. (2024) · International Journal of Applied Earth Observation and Geoinformation · doi:10.1016/j.jag.2024.104085 · Cited by 25
Ponomarenko, S., McLennan, D., Pouliot, D., & Wagner, J. (2019) · Canadian Journal of Remote Sensing · doi:10.1080/07038992.2019.1682980 · Cited by 25
“The interpreter also alleviates problems related to sample imbalances known to effect machine learning approaches (Buda et al. 2018; Pouliot et al. 2016; Wang et al. 2016).”
BackgroundThresholding Strategies for Deep Learning with Highly Imbalanced Big Data
Johnson, J. M., & Khoshgoftaar, T. M. (2020) · Advances in intelligent systems and computing · doi:10.1007/978-981-15-6759-9_9 · Cited by 26
Attention-Based Network for Weak Labels in Neonatal Seizure Detection.
Isaev, D., Tchapyjnikov, D., Cotten, C. M., Tanaka, D. T., Martínez, N., Bertrán, M., Sapiro, G., & Carlson, D. (2020) · PubMed · Cited by 25
The effects of sample size and sample prevalence on cellular automata simulation of urban growth
Zhang, B., & Xia, C. (2021) · International Journal of Geographical Information Systems · doi:10.1080/13658816.2021.1931237 · Cited by 25
“This trend is significant in the variations of the simulation results of the LR-CA and SVM-CA models, but is not significant in the results of the ANN-CA models, which may be due to the perturbation of the uncertainty in the ANN training (Buda et al. 2018).”
BackgroundResultTan, W., Guan, P., Wu, L., Chen, H., Li, J., Ling, Y., Fan, T., Wang, Y., Li, J., & Yan, B. (2021) · Annals of Translational Medicine · doi:10.21037/atm-21-1171 · Cited by 24
“Class imbalance typically poses a great challenge to designing model structures (25,28), updating model parameters for convergence, and maintaining the balance of test performance for each category.”
BackgroundHe, C., Wang, R., & Chen, X. (2021) · IEEE/CVF Conference on Computer Vision and Pattern Recognition Workshops (CVPRW) · doi:10.1109/cvprw53098.2021.00395 · Cited by 26
“Class imbalance or long-tail distribution is ubiquitous, and the battle against it has a long history [26, 64, 4].”
BackgroundMethodRandhawa, J. K., Chiang, M., Porporato, N., Pardeshi, A. A., Dredge, J., Apolo, G., Tun, T. A., Quah, J. H. M., Tan, M. C. J., Higashita, R., et al. (2021) · British Journal of Ophthalmology · doi:10.1136/bjophthalmol-2021-319470 · Cited by 26
“33 While our deep learning classifier was originally developed using population-based data, the CHES training and test datasets were balanced based on the number of open and closed angles to avoid introducing biases in the detection of angle closure.”
MethodDenissen, S., Chén, O. Y., Mey, J. D., Vos, M. D., Schependom, J. V., Sima, D. M., & Nagels, G. (2021) · Journal of Personalized Medicine · doi:10.3390/jpm11121349 · Cited by 27
“Like overfitting, it can lead to the poor generalization of an algorithm [26].”
BackgroundMethodKe, M., Xu, N., Zhang, Z., Qiu, D., Kang, J., Lu, T., Wang, T., Peijnenburg, W. J. G. M., Sun, L., Hu, B., et al. (2022) · Environmental Microbiology · doi:10.1111/1462-2920.16175 · Cited by 27
Diagnosis of heart failure from imbalance datasets using multi-level classification
Li, D., Zheng, C., Zhao, J., & Liu, Y. (2022) · Biomedical Signal Processing and Control · doi:10.1016/j.bspc.2022.104538 · Cited by 25
Amending Facial Expression Representation via De-albino
Shi, J., Zhu, S., & Liang, Z. (2022) · Chinese Control Conference (CCC) · doi:10.23919/ccc55666.2022.9901738 · Cited by 25
“Ascribe to the learning bias of CNN, neurons will give more weight to categories that are easy to differentiate or contain a large number of examples [26].”
BackgroundRe-Thinking Federated Active Learning Based on Inter-Class Diversity
Kim, S., Bae, S., Song, H., & Yun, S.-Y. (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/cvpr52729.2023.00384 · Cited by 25
“Besides, ρ indicator is the ratio of class imbalance for the aggregated global data of all clients [7, 21].”
BackgroundWeighted Focal Loss: An Effective Loss Function to Overcome Unbalance Problem of Chest X-ray14
Qin, R., Qiao, K., Wang, L., Zeng, L., Chen, J., & Yan, B. (2018) · IOP Conference Series Materials Science and Engineering · doi:10.1088/1757-899x/428/1/012022 · Cited by 27
“There are many methods to deal with unbalance-class problem of database [7].”
MethodProgressive Improved Convolutional Neural Network for Avionics Fault Diagnosis
Chen, S., Ge, H., Li, J., & PECHT, M. G. (2019) · IEEE Access · doi:10.1109/access.2019.2954170 · Cited by 27
“On this basis, the developed progressive SDCNN is then evaluated and compared with other CNNs under samples with noise.”
BackgroundMethodInfluentialLibrenza‐Garcia, D., Passos, I. C., Feiten, J. G., Lotufo, P. A., Goulart, A. C., Santos, I. S., Viana, M. C., Benseñor, I. M., & Brunoni, A. R. (2020) · Psychological Medicine · doi:10.1017/s0033291720001579 · Cited by 25
“Therefore, when dealing with unbalanced datasets, AUC and balanced accuracies are more proper performance metrics to evaluate a machine learning classifier than the traditional accuracy values (Buda, Maki, & Mazurowski, 2018; Luque, Carrasco, Martín, & de las Heras, 2019).”
Calibrated Resampling for Imbalanced and Long-Tails in Deep Learning
Bellinger, C., Corizzo, R., & Japkowicz, N. (2021) · Lecture notes in computer science · doi:10.1007/978-3-030-88942-5_19 · Cited by 26
Deep Learning for Scene Classification: A SurveyPreprint
Zeng, D., Liao, M., Tavakolian, M., Guo, Y., Zhou, B., Hu, D., Pietikäinen, M., & Liu, L. (2021) · arXiv (Cornell University) · doi:10.48550/arxiv.2101.10531 · Cited by 44
“Therefore, learning a model which respects both type of categories and equally performs well on frequent and infrequent ones remains a great challenge and needs further exploration [254], [255], [256].”
BackgroundA Novel IDS Based on Jaya Optimizer and Smote-ENN for Cyberattacks Detection
Albasheer, F. O., Haibatti, R. R., Agarwal, M., & Nam, S. Y. (2024) · IEEE Access · doi:10.1109/access.2024.3431534 · Cited by 24
“Research conducted in [22] has demonstrated that oversampling is the most effective approach for addressing imbalances in deep learning and does not lead to overfitting of the model.”
BackgroundOpportunities and Challenges in Applying AI to Evolutionary Morphology
He, Y., Mulqueeney, J. M., Watt, E. C., Salili‐James, A., Barber, N. S., Camaiti, M., Hunt, E. S. E., Kippax-Chui, O., Knapp, A., Lanzetti, A., et al. (2024) · Integrative Organismal Biology · doi:10.1093/iob/obae036 · Cited by 32
Aubaidan, B. H., Kadir, R. A., Lajb, M. T., Anwar, M., Qureshi, K. N., Taha, B. A., & Ghafoor, K. Z. (2025) · Intelligent Data Analysis · doi:10.1177/1088467x241305509 · Cited by 28
Quality control of online monitoring data of air pollutants using artificial neural networks
Wang, Z., Feng, J.-J., Fu, Q., Gao, S., Chen, X., & Cheng, J. (2019) · Air Quality Atmosphere & Health · doi:10.1007/s11869-019-00734-4 · Cited by 24
“Because the dataset was highly imbalanced (invalid observations accounted for ~4% of the total observations), oversampling was used, i.e., samples from invalid observations were randomly selected and replicated to balance the positive and negative samples (Buda et al. 2018).”
MethodClass Balanced Loss for Image Classification
Wang, L., Wang, C., Sun, Z., Cheng, S., & Guo, L. (2020) · IEEE Access · doi:10.1109/access.2020.2991237 · Cited by 28
Bibliometric Analysis of the Application of Convolutional Neural Network in Computer Vision
Huie, C., & Deng, Z. (2020) · IEEE Access · doi:10.1109/access.2020.3019336 · Cited by 25
“Mateusz B uses three benchmark datasets (MNIST, CIFAR-10, and ImageNet) to continuously reduce the complexity of and study the impact of imbalance on classification performance [44].”
MethodHu, X., Zhang, H., Ma, D., Wang, R., & Zheng, J. (2020) · Neurocomputing · doi:10.1016/j.neucom.2020.11.009 · Cited by 24
PLM: Partial Label Masking for Imbalanced Multi-label Classification
Duarte, K., Rawat, Y. S., & Shah, M. A. (2021) · IEEE/CVF Conference on Computer Vision and Pattern Recognition Workshops (CVPRW) · doi:10.1109/cvprw53098.2021.00308 · Cited by 23
“Extensive overviews of the problem can be found for both classical methods [20, 15] and convolutional neural networks [2].”
BackgroundTATL: Task agnostic transfer learning for skin attributes detection
Nguyen, M.-D., Nguyen, T., Vu, H., Pham, Q., Nguyen, M.-D., Nguyen, B. T., & Sonntag, D. (2022) · Medical Image Analysis · doi:10.1016/j.media.2022.102359 · Cited by 24
“Moreover, we apply strategy one class one model (Buda et al., 2018), thus each classifier Target-Segmenter in TATL only detects one attribute, which alleviates the data’s imbalance.”
MethodSelf-training Guided Adversarial Domain Adaptation For Thermal Imagery
Akkaya, İ. B., Altinel, F., & Halıcı, U. (2021) · IEEE/CVF Conference on Computer Vision and Pattern Recognition Workshops (CVPRW) · doi:10.1109/cvprw53098.2021.00488 · Cited by 24
“Effective classification for imbalanced data is an important field of research since class imbalance exists in many real-world applications [3, 18].”
BackgroundMosser, L., & Naeini, E. Z. (2022) · Geophysics · doi:10.1190/geo2021-0318.1 · Cited by 25
Using artificial intelligence to find design errors in the engineering drawings
Джусупова, Р., Banotra, R., Bosch, J., & Olsson, H. H. (2023) · Journal of Software Evolution and Process · doi:10.1002/smr.2543 · Cited by 24
“The network, which used the Mish activation function, 41 was trained with a learning rate of 0.001 with the max number of batches being 16,000, where each batch size is equal to 64 training examples, and 12,800, and 14,400 steps.”
BackgroundMethodInfluentialA Comparative Approach to Threshold Optimization for Classifying Imbalanced Data
Hancock, J., Johnson, J. M., & Khoshgoftaar, T. M. (2022) · IEEE International Conference on Collaboration and Internet Computing (CIC) · doi:10.1109/cic56439.2022.00028 · Cited by 24
“Buda et al. [12] explore the importance of output thresh-olding using three popular image data sets and convolutional neural networks.”
MethodGenerative Table Pre-training Empowers Models for Tabular Prediction
Zhang, T., Wang, S., Yan, S., Li, J., & Liu, Q. (2023) · Conference on Empirical Methods in Natural Language Processing (EMNLP) · doi:10.18653/v1/2023.emnlp-main.917 · Cited by 49
Rotemberg, V. M., Halpern, A. C., Dusza, S., & Codella, N. (2019) · Seminars in Cutaneous Medicine and Surgery · doi:10.12788/j.sder.2019.013 · Cited by 24
“13,25”
Multiple Fake Classes GAN for Data Augmentation in Face Image Dataset
Ali-Gombe, A., Elyan, E., & Jayne, C. (2019) · International Joint Conference on Neural Networks (IJCNN) · doi:10.1109/ijcnn.2019.8851953 · Cited by 24
“of data which may be caused by under-sampling or possible overfitting due to oversampling [4].”
BackgroundDeep Learning and Thresholding with Class-Imbalanced Big Data
Johnson, J. M., & Khoshgoftaar, T. M. (2019) · International Conference on Machine Learning and Applications · doi:10.1109/icmla.2019.00134 · Cited by 24
“These existing studies address class imbalance with data sampling [18], [19], [20], loss functions that balance class-wise weight updates [21], [22], [23], or complex hybrid methods that leverage deep representation learning [24], [25], [26].”
BackgroundMattheuwsen, L., & Vergauwen, M. (2020) · Remote Sensing · doi:10.3390/rs12223820 · Cited by 24
Yang, L., Wang, D., Hu, J., & Zhu, G. (2023) · Rock Mechanics and Rock Engineering · doi:10.1007/s00603-023-03623-6 · Cited by 23
“If not processed, this phenomenon of data imbalance will result in different learning effects of CNN on different types of rock fragments under different geological conditions (Buda et al. 2018; Chi and Wang 2022).”
Xu, B., Mao, Y., Wang, W., & Chen, G. (2024) · Frontiers in Artificial Intelligence · doi:10.3389/frai.2024.1299169 · Cited by 27
Liu, Y., Sun, T., Wu, K., Xiang, W., Zhang, J., Zhang, H., & Feng, M. (2025) · Natural Resources Research · doi:10.1007/s11053-025-10462-5 · Cited by 27
“SMOTE is a statistical technique for increasing the number of minority samples in a dataset by creating synthetic samples (Chawla et al., 2002; Buda et al., 2018).”
Ss-InfoGAN for Class-Imbalance Classification of Bearing Faults
Wu, J., Zhao, Z., Sun, C., Yan, R., & Chen, X. (2020) · Procedia Manufacturing · doi:10.1016/j.promfg.2020.07.003 · Cited by 23
Mathur, P., Kakwani, K., Diplav, Kudavelly, S. R., & Rama Raju, G. A. (2020) · Annual International Conference of the IEEE Engineering in Medicine and Biology Society (EMBC) · doi:10.1109/embc44109.2020.9176703 · Cited by 23
“tackle the class imbalance problem [12] when the #foreground pixels << #background pixels.”
BackgroundQian, L., Lv, Z., Zhang, K., Wang, K., Zhu, Q., Zhou, S., Chang, C., & Tian, J. (2021) · Annals of Translational Medicine · doi:10.21037/atm-20-3981 · Cited by 23
“For optimal performance of the convolution network model, the sample sizes in the two groups (upstaged and pure DCIS) should be equal (11).”
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
