Experimental perspectives on learning from imbalanced data

By Jason Van Hulse, Taghi M. Khoshgoftaar, Amri Napolitano. International Conference on Machine Learning, 2007

842 citations recorded by the source corpus.

Subfields: Artificial Intelligence, Electrical and Electronic Engineering

View Paper: https://www.semanticscholar.org/paper/f26bb9a2034b8fb219b5d03546fa71604642b3fd

How researchers see this paper

The researchers whose trusted networks rank it highest, strongest advocate first. Every researcher gets a different order.

In Artificial Intelligence

In Electrical and Electronic Engineering

Citations in the corpus

Combating the Small Sample Class Imbalance Problem Using Feature Selection (2010, methodology): “Random undersampling has been shown to be one of the best resampling algorithms [49], and SMOTE alleviates the small sample problem by doubling the minority class size.”
Data Distributional Properties Drive Emergent Few-Shot Learning in Transformers (background): “While non-uniformity can impair both supervised and reinforcement learning (Chan et al., 2022; Van Hulse et al., 2007), we show here that non-uniform training distributions can lead to the emergence of at least one useful and interesting capability, and thus can be an opportunity as well as a…”
Comparing Boosting and Bagging Techniques With Noisy and Imbalanced Data (2011, methodology): “Data sampling [3], [4], a common method for dealing with class imbalance, balances the class distribution (CD) of the training data by either adding examples to the minority class (oversampling) or removing examples from the majority class (undersampling).”
RUSBoost: A Hybrid Approach to Alleviating Class Imbalance (2010, methodology): “RUSBoost applies RUS, which is a technique that randomly removes examples from the majority class.”
Learning Entailment Relations by Global Graph Structure Optimization (2012, positive): “Last, we use a balanced number of positive and negative examples, because classifiers tend to perform poorly on the minority class when trained on imbalanced data (Van Hulse, Khoshgoftaar, and Napolitano 2007 ; Nikulin 2008 ).”
An instance level analysis of data complexity (2014, background): “Keywords Instance hardness · Dataset hardness · Data complexity”
Data mining for credit card fraud: A comparative study (2011, background): “A recent experimental study of various sampling procedures used with different learning algorithms [25] found performance of sampling techniques to vary with learning algorithm used, and also with respect to performance measures.”
Cost-sensitive learning methods for imbalanced data (2010, methodology): “Moreover, [26] applied synthetic minority oversampling technique (SMOTE [4]) to balance the dataset first, then built the model using SVM with different costs proposed by [13]; [27] and [28] applied some common classifiers (e.”
Addressing Class Imbalance in Federated Learning (2020, methodology): “Data-level approaches leverage data re-sampling (Van Hulse, Khoshgoftaar, and Napolitano 2007; Mani and Zhang 2003) and data augmentation (Lee, Park, and Kim 2016; Pouyanfar et al. 2018).”
SMOTE for high-dimensional class-imbalanced data (2013, methodology): “However, it was recently experimentally observed using low-dimensional data that simple undersampling tends to outperform SMOTE in most situations [8].”
Class Imbalance, Redux (2011, background): “This work thus provides an explanation for the otherwise surprising observation that undersampling tends often to outperform what are ostensibly more advanced techniques (e.g., SMOTE) [11].”
A Survey of Predictive Modeling on Imbalanced Domains (2016, background): “We must also emphasize that using different evaluation metrics may lead to different conclusions (e.g., Van Hulse et al. [2007]), which is problematic and reinforces the need for ﬁnding suitable metrics that are capable of assessing correctly the user goals.”
Data Distributional Properties Drive Emergent In-Context Learning in Transformers (2022, background): “While it can impair both supervised and reinforcement learning (Chan et al., 2022; Van Hulse et al., 2007), we show here that non-uniform training distributions can induce the emergence of at least one useful and interesting capability, and thus can be an opportunity as well as a challenge.”
Online Defect Prediction for Imbalanced Data (2015, background): “Classifying imbalanced data is an open challenge in the machine learning community [13, 45, 56].”
The Constrained Weight Space SVM: Learning with Ranked Features (2011, background): “In this case, we have 15 positive terms di- 5Random undersampling of the majority class has been shown to mitigate class imbalance (Van Hulse et al., 2007). vided into 3 ranked term classes and 7 negative terms divided into 2 ranked classes (again derived independently).”