A Survey on Small Sample Imbalanced Problem

This repository provides a brief summary of methods from our review paper: A Survey on Small Sample Imbalanced Problem: Metrics, Feature Analysis, and Solutions.

The small sample imbalance (S&I) problem is a major challenge in machine learning and data analysis. It is characterized by a small number of samples and an imbalanced distribution of classes, which leads to poor model performance. In addition, feature distribution and overlapping noise in the data significantly impact the classification performance. We emphasize a detailed analysis of the dataset before developing a matching solution. We first summarize imbalance metrics and complexity analysis methods that provide interpretable benchmarks for S&I. Second, we review the latest solutions for conventional, data complexity-based, and extreme S&I problems, thereby providing insights for future research.

See our paper for more details.

Imbalance Measurement

Imbalance Metrics Based on Sample Distribution

• (ID): Measuring the class imbalance extent of multi-class problems. [paper]

• (LRID): LRID: A new metric of multi-class imbalance degree based on likelihood-ratio test. [paper]

• (Imbalance Coefficient): The impact of class imbalance in classification performance metrics based on the binary confusion matrix. [paper]

• (RIR): TF-IGM revisited: Imbalance text classification with relative imbalance ratio. [paper]

Measuring Imbalance with Classification Performance

• (Augmented R-value): Dealing with overlap and imbalance: a new metric and approach. [paper]

• (adjustedIR): Adjusting the imbalance ratio by the dimensionality of imbalanced data. [paper]

• (IBI3, BI3): Bayes Imbalance Impact Index: A Measure of Class Imbalanced Data Set for Classification Problem. [paper]

• (IF): Imbalance factor: a simple new scale for measuring inter-class imbalance extent in classification problems. [paper]

• (MIR, MiCIR): Purposive Data Augmentation Strategy and Lightweight Classification Model for Small Sample Industrial Defect Dataset. [paper]

Data Complexity Measurement

• Complexity measures of supervised classification problems. [paper]

• Analysis of data complexity measures for classification. [paper]

• How Complex Is Your Classification Problem?: A Survey on Measuring Classification Complexity. [paper] [code]

• Assessing the data complexity of imbalanced datasets. [paper] [results]

• On the joint-effect of class imbalance and overlap: a critical review. [paper] [code: pycol] [code: open-source-imbalance-overlap]

• A unifying view of class overlap and imbalance: Key concepts, multiview panorama, and open avenues for research. [paper] [code: dcol] [code: ImbCoL]

Conventional S&I Solutions

Data-level

Resampling

• Synthetic Oversampling

  • (SMOTE): SMOTE: synthetic minority over-sampling technique. [paper] [code]

  • (Borderline-SMOTE): Borderline-SMOTE: A New Over-Sampling Method in Imbalanced Data Sets Learning. [paper]

  • (ADASYN): ADASYN: Adaptive synthetic sampling approach for imbalanced learning. [paper] [code]

  • (WK-SMOTE): Classification of Imbalanced Data by Oversampling in Kernel Space of Support Vector Machines. [paper]

  • (MC-SMOTE): Imbalanced Classification Based on Minority Clustering Synthetic Minority Oversampling Technique With Wind Turbine Fault Detection Application. [paper]

• Combined Approaches

  • (SUNDO): A method for resampling imbalanced datasets in binary classification tasks for real-world problems. [paper]

  • (WRO): An Improved Algorithm for Imbalanced Data and Small Sample Size Classification. [paper]

Data augmentation

• Traditional method

  • An imbalanced small sample slab defect recognition method based on image generation. [paper]

• VAE-based

  • Variational autoencoder based synthetic data generation for imbalanced learning. [paper]

  • (WM-CVAE): A novel data augmentation approach to fault diagnosis with class-imbalance problem. [paper]

  • (TL-VAE): Gas Path Fault Diagnosis Method Under Small Samples With Inter-Class Imbalance. [paper]

• GAN-based

  • (DAC): Data Augmentation Classifier for Imbalanced Fault Classification. [paper]

  • Using CGAN to Deal with Class Imbalance and Small Sample Size in Cybersecurity Problems. [paper] [code: cgan]

  • (ACGAN-SN): A Novel Method for Fault Diagnosis of Bearings with Small and Imbalanced Data Based on Generative Adversarial Networks. [paper]

  • (SCA-GAN): SCA-CGAN: A New Side-Channel Attack Method for Imbalanced Small Samples. [paper]

  • (AGMAN): Auxiliary generative mutual adversarial networks for class-imbalanced fault diagnosis under small samples. [paper]

  • (ConvGeN): ConvGeN: A convex space learning approach for deep-generative oversampling and imbalanced classification of small tabular datasets. [paper] [code: ConvGeN] [code: NextConvGeN]

• Diffusion Model

  • A Novel Data Augmentation Method Based on Denoising Diffusion Probabilistic Model for Fault Diagnosis Under Imbalanced Data. [paper] [code: DDPM]

Algorithm-level

Deep learning model

• (DSLWCN-VAFL): Lightweight Network with Variable Asymmetric Rebalancing Strategy for Small and Imbalanced Fault Diagnosis. [paper]

• Transformer-Based Disease Identification for Small-Scale Imbalanced Capsule Endoscopy Dataset. [paper]

• Wind Turbine Fault Diagnosis for Class-Imbalance and Small-Size Data Based on Stacked Capsule Autoencoder. [paper]

Cost-sensitive learning

• Class-Wise Difficulty-Balanced Loss for Solving Class-Imbalance. [paper] [code]

Ensemble learning

• A novel ensemble method for classifying imbalanced data. [paper]

Transfer learning

• (Rare-Transfer): Transfer learning for class imbalance problems with inadequate data. [paper]

• (CIATL): Class-Imbalance Adversarial Transfer Learning Network for Cross-Domain Fault Diagnosis With Imbalanced Data. [paper]

• Edge–Cloud Collaboration Detection Approach for Small-Sample Imbalanced Faults in Power Lines. [paper]

• (G-TELM): Generalized Transfer Extreme Learning Machine for Unsupervised Cross-Domain Fault Diagnosis With Small and Imbalanced Samples. [paper]

Few-shot learning

• (PTAFedIF): Privacy-Preserving Incipient Fault Identification in Distribution Networks Under Small Sample and Imbalanced Data Distribution Conditions. [paper]

• (TRN): Transfer Relation Network for Fault Diagnosis of Rotating Machinery With Small Data. [paper]

Hybrid solutions

• Early Fault Detection in Induction Motors Using AdaBoost With Imbalanced Small Data and Optimized Sampling. [paper]

• Exploiting Synthetically Generated Data with Semi-Supervised Learning for Small and Imbalanced Datasets. [paper]

• A diagnosis method for imbalanced bearing data based on improved SMOTE model combined with CNN-AM. [paper]

• An oversampling method for wafer map defect pattern classification considering small and imbalanced data. [paper]

• DeepSMOTE: Fusing Deep Learning and SMOTE for Imbalanced Data. [paper] [code]

• Fault Diagnosis of Rolling Bearing with Imbalanced Small Sample Scenarios. [paper]

• Deep learning approach for defective spot welds classification using small and class-imbalanced Datasets. [paper] [code]

• Single Model Deep Learning on Imbalanced Small Datasets for Skin Lesion Classification. [paper] [code]

• Improved Generative Adversarial Network for Bearing Fault Diagnosis with a Small Number of Data and Unbalanced Data. [paper]

• Imbalanced fault diagnosis of rotating machinery via multi-domain feature extraction and cost-sensitive learning. [paper]

• Convolution-GRU Based on Independent Component Analysis for fMRI Analysis with Small and Imbalanced Samples. [paper]

• Automated Traffic Incident Detection: Coping With Imbalanced and Small Datasets. [paper]

• Pedestrian gender classification on imbalanced and small sample datasets using deep and traditional features. [paper]

• A novel building heat pump system semi-supervised fault detection and diagnosis method under small and imbalanced data. [paper]

• Classification of autonomous vehicle crash severity: Solving the problems of imbalanced datasets and small sample size. [paper]

Data Complexity Solutions for S&I

Data-level

Resampling

• Filtering-Based

  • (SMOTE-Tomek, SMOTE-ENN): A study of the behavior of several methods for balancing machine learning training data. [paper] [code: SMOTE_TomekLinks] [code: SMOTE-ENN]

  • (SMOTE-IPF): SMOTE–IPF: Addressing the noisy and borderline examples problem in imbalanced classification by a re-sampling method with filtering. [paper]

  • (SMOTE-WENN): SMOTE-WENN: Solving class imbalance and small sample problems by oversampling and distance scaling. [paper]

• Feature-Based

  • (FW-SMOTE): FW-SMOTE: A feature-weighted oversampling approach for imbalanced classification. [paper]

• Sample Concatenation

  • (Re-SC): Resampling algorithms based on sample concatenation for imbalance learning. [paper]

• Clustering + Weighting

  • (A-SUWO): Adaptive semi-unsupervised weighted oversampling (A-SUWO) for imbalanced datasets. [paper]

  • (IA-SUWO): IA-SUWO: An Improving Adaptive semi-unsupervised weighted oversampling for imbalanced classification problems. [paper]

  • (MWMOTE): MWMOTE–Majority Weighted Minority Oversampling Technique for Imbalanced Data Set Learning [paper] [code]

  • (NI-MWMOTE): NI-MWMOTE: An improving noise-immunity majority weighted minority oversampling technique for imbalanced classification problems. [paper]

  • (IMWMOTE): IMWMOTE: A novel oversampling technique for fault diagnosis in heterogeneous imbalanced data. [paper]

  • (EM-LWMOTE): A Local Adaptive Minority Selection and Oversampling Method for Class-Imbalanced Fault Diagnostics in Industrial Systems. [paper]

• Cleaning-Based

  • (CCR): CCR: A combined cleaning and resampling algorithm for imbalanced data classification. [paper] [code]

  • (MC-CCR): Combined cleaning and resampling algorithm for multi-class imbalanced data with label noise. [paper] [code]

  • (RB-CCR): RB-CCR: Radial-Based Combined Cleaning and Resampling algorithm for imbalanced data classification. [paper] [code]

• Radial-Based

  • (RBO): Radial-Based oversampling for noisy imbalanced data classification. [paper] [code]

  • (MC-RBO): Radial-Based Oversampling for Multiclass Imbalanced Data Classification. [paper] [code]

• Distance-based

  • (MDO): To Combat Multi-Class Imbalanced Problems by Means of Over-Sampling Techniques. [paper]

  • (AMDO): AMDO: An Over-Sampling Technique for Multi-Class Imbalanced Problems. [paper]

Data augmentation

• (MoGAN): Oversampling adversarial network for class-imbalanced fault diagnosis. [paper]

• (FAWGAN-GN): Full Attention Wasserstein GAN With Gradient Normalization for Fault Diagnosis Under Imbalanced Data. [paper]

• (PCWGAN-GP): PCWGAN-GP: A New Method for Imbalanced Fault Diagnosis of Machines. [paper]

• (CFGAN): Categorical Feature GAN for Imbalanced Intelligent Fault Diagnosis of Rotating Machinery. [paper]

Feature-level

• (S2N and FAST): Combating the Small Sample Class Imbalance Problem Using Feature Selection. [paper]

• (AMF): Feature Selection Metric Using AUC Margin for Small Samples and Imbalanced Data Classification Problems. [paper]

• (DBFS): DBFS: An effective Density Based Feature Selection scheme for small sample size and high dimensional imbalanced data sets. [paper]

• (UIG-CFGVM): A Hybrid Feature Selection Algorithm Applied to High-dimensional Imbalanced Small-sample Data Classification. [paper]

• (MOSNS and MOSS): Feature selection and classification by minimizing overlap degree for class-imbalanced data in metabolomics. [paper]

Algorithm-level

• (Soft-Hybrid): Improving classification rate constrained to imbalanced data between overlapped and non-overlapped regions by hybrid algorithms. [paper]

• (Meta-learning): Learning to Reweight Examples for Robust Deep Learning. [paper]

• (Class-Balanced loss): Class-balanced loss based on effective number of samples. [paper] [code]

• (RBBag): Roughly balanced bagging for imbalanced data. [paper]

• (RBBag+RSM and MRBBag): Multi-class and feature selection extensions of Roughly Balanced Bagging for imbalanced data. [paper]

Hybrid solutions

• (WMODA): Imbalanced data fault diagnosis of rotating machinery using synthetic oversampling and feature learning [paper]

• (PCA-SMOTE-SVM): Chiller fault diagnosis with field sensors using the technology of imbalanced data. [paper]

• Combining feature selection and hybrid approach redefinition in handling class imbalance and overlapping for multi-class imbalanced. [paper]

• A Weakly Supervised Learning-Based Oversampling Framework for Class-Imbalanced Fault Diagnosis. [paper] [code]

• An Ensemble Resampling Based Transfer AdaBoost Algorithm for Small Sample Credit Classification with Class Imbalance. [paper]

• New imbalanced fault diagnosis framework based on Cluster-MWMOTE and MFO-optimized LS-SVM using limited and complex bearing data. [paper]

• New imbalanced bearing fault diagnosis method based on Sample-characteristic Oversampling TechniquE (SCOTE) and multi-class LS-SVM. [paper]

• Novel extended NI-MWMOTE-based fault diagnosis method for data-limited and noise-imbalanced scenarios. [paper]

Extreme S&I Solutions

• (SWIM): Synthetic Oversampling with the Majority Class: A New Perspective on Handling Extreme Imbalance. [paper]

• (Few-shot GAN): Few-Shot GAN: Improving the Performance of Intelligent Fault Diagnosis in Severe Data Imbalance. [paper]

• (UCML): Multiset Feature Learning for Highly Imbalanced Data Classification. [paper]

• (DSADRSViT-IIRL): Residual Shrinkage ViT with Discriminative Rebalancing Strategy for Small and Imbalanced Fault Diagnosis. [paper]

Long-tail distribution

• (IFL): Invariant Feature Learning for Generalized Long-Tailed Classification. [paper] [code]

• (RCAL): When Noisy Labels Meet Long Tail Dilemmas: A Representation Calibration Method. [paper]

• (ProLT): Revealing the Proximate Long-Tail Distribution in Compositional Zero-Shot Learning. [paper]

• (DO): Decoupled Optimisation for Long-Tailed Visual Recognition. [paper]

• (ProCo): Probabilistic Contrastive Learning for Long-Tailed Visual Recognition. [paper] [code]

Contact

If you have any suggestions or find our work helpful, feel free to contact us.

Email: {guijie,zhaosxian}@seu.edu.cn