The effect of class distribution on classifier learning: an empirical study

By Gary M. Weiss, Foster J. Provost. 2001

531 citations recorded by the source corpus.

Subfields: Accounting, Artificial Intelligence, Information Systems

View Paper: https://www.semanticscholar.org/paper/31789bf4a7fb4b4024648269aa4335e286de5780

How researchers see this paper

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

In Accounting

In Artificial Intelligence

In Information Systems

Citations in the corpus

Churn prediction in subscription services: An application of support vector machines while comparing two parameter-selection techniques (2008, background): “So it is not uncommon to train a model with a non-natural distribution (Chan & Stolfo, 1998; Weiss & Provost, 2001).”
Training neural network classifiers for medical decision making: The effects of imbalanced datasets on classification performance (2008, background): “In another study (Weiss & Provost, 2001), the authors evaluated the effect of the extent of data imbalance on classifier performance.”
Calibrating Probability with Undersampling for Unbalanced Classification (2015, background): “Even if undersampling produces poorly calibrated probability estimates p̂s, several studies have shown that it often provides better predictive accuracy than p̂ [25], [14].”
Class Imbalances versus Class Overlapping: An Analysis of a Learning System Behavior (2004, background): “In the last years, several works have been published in the machine learning literature aiming to overcome the class imbalance problem [7, 12].”
A Comparative Performance Analysis of Data Resampling Methods on Imbalance Medical Data (2021, unknown): “rectly identified as minority samples [4], [5].”
Balancing Training Data for Automated Annotation of Keywords: a Case Study (2003, background): “This decision is motivated by the results presented in [11], in which is shown that allocating 50% of the training examples to the minority class, while it does not always yield optimal results, it generally leads to results which are no worse than, and often superior to, those which use the natural”
Class-Boundary Alignment for Imbalanced Dataset Learning (2003, background): “Over-sampling (Chawla et al., 2000; Weiss & Provost, 2001) is the opposite of the under-sampling approach.”
A balanced accuracy function for epistasis modeling in imbalanced datasets using multifactor dimensionality reduction (2007, methodology): “To minimize the effect that training set size has on classifier performance it is important to choose the training data carefully; which involves choosing an appropriate class distribution in the training dataset [Weiss and Provost, 2001].”
To Combat Multi-Class Imbalanced Problems by Means of Over-Sampling Techniques (2016, methodology): “It is a simple method, while it has been argued that exact copies of the minority class instances and very specific rules that this kind of over-sampling generates can lead the algorithm to overfit and over-generalize [41], [53].”
Optimizing Classifier Performance via an Approximation to the Wilcoxon-Mann-Whitney Statistic (2003, methodology): “Weiss & Provost (2001) report that the best weight (class distribution ) needs to be experimentally determined for each data set.”
C4.5, Class Imbalance, and Cost Sensitivity: Why Under-Sampling beats Over-Sampling (2003, background): “Cost sensitivity can also be introduced by changing the threshold at the leaves of the tree, allowing the data to be used without sampling ( Weiss & Provost, 2001 ).”
Training and assessing classification rules with imbalanced data (2012, background): “It has been largely emphasized (He and Garcia 2009; Weiss and Provost 2001; Weiss 2004) that the use of common performance measures, such as the error rate, may yield to misleading results because they strongly depend on the class distribution.”
Learning When Training Data are Costly: The Effect of Class Distribution on Tree Induction (2003, methodology): “This method (Weiss & Provost, 2001) is justified informally, using a simple, intuitive, argument.”
Adaptive Machine Learning for Credit Card Fraud Detection (2015, methodology): “In practice, sampling methods are often used to balance datasets with skewed class distributions because several classifiers have empirically shown better performance when trained on balanced dataset [88, 90].”
On learning algorithm selection for classification (2006, methodology): “Therefore, this method is criticised as an impractical technique [9].”
Learning from Imbalanced Data (2009, background): “Studies have shown that for several base classifiers, a balanced data set provides improved overall classification performance compared to an imbalanced data set [35], [36], [37].”