Learning with limited minority class data

Khoshgoftaar, Taghi M.; Seiffert, Chris; Hulse, Jason Van; Napolitano, Amri; Folleco, Andres

doi:10.1109/icmla.2007.76

Cited by 83 publications

(49 citation statements)

References 11 publications

Supporting

Mentioning

Contrasting

Unclassified

Order By: Relevance

“…obtaining the same cardinality in classes) is not optimal when dealing with such rare classes. For instance, the reader can consult the comprehensive study with many data sets and classifiers showing that depending on combination of data and classifiers the ratios of modified majority vs. minority class cardinalities like 3:1 and 2:1 quite often outperformed the most popular ratio 1:1 [33] Therefore, researchers proposed more elaborated methods that attempt at taking into account data characteristics and factors influencing nature of class imbalance.…”

Section: Improving Classifiers By Focused Re-sampling Methodsmentioning

confidence: 99%

Overlapping, Rare Examples and Class Decomposition in Learning Classifiers from Imbalanced Data

Stefanowski

2013

Smart Innovation, Systems and Technologies

View full text Add to dashboard Cite

This paper deals with inducing classifiers from imbalanced data, where one class (a minority class) is under-represented in comparison to the remaining classes (majority classes). The minority class is usually of primary interest and it is required to recognize its members as accurately as possible. Class imbalance constitutes a difficulty for most algorithms learning classifiers as they are biased toward the majority classes. The first part of this study is devoted to discussing main properties of data that cause this difficulty. Following the review of earlier, related research several types of artificial, imbalanced data sets affected by critical factors have been generated. The decision trees and rule based classifiers have been generated from these data sets. Results of first experiments show that too small number of examples from the minority class is not the main source of difficulties. These results confirm the initial hypothesis saying the degradation of classification performance is more related to the minority class decomposition into small sub-parts. Another critical factor concerns presence of a relatively large number of borderline examples from the minority class in the overlapping region between classes, in particular for non-linear decision boundaries. The novel observation is showing the impact of rare examples from the minority class located inside the majority class. The experiments make visible that stepwise increasing the number of borderline and rare examples in the minority class has larger influence on the considered classifiers than increasing the decomposition of this class. The second part of this paper is devoted to studying an improvement of classifiers by pre-processing of such data with resampling methods. Next experiments examine the influence of the identified critical data factors on performance of 4 different pre-processing re-sampling methods: two versions of random over-sampling, focused under-sampling NCR and the hybrid method SPIDER. Results show that if data is sufficiently disturbed by borderline and rare examples SPIDER and partly NCR work better than over-sampling.

show abstract

Section: Improving Classifiers By Focused Re-sampling Methodsmentioning

confidence: 99%

Overlapping, Rare Examples and Class Decomposition in Learning Classifiers from Imbalanced Data

Stefanowski

2013

Smart Innovation, Systems and Technologies

View full text Add to dashboard Cite

show abstract

“…Another feature of oversampling to consider is the class distribution ratio. Khoshgoftaar et al (2007) reported that an even distribution is not always optimal when dealing with data rarity. To ensure a more representative minority class, clusters could be identified in the minority class from which to sample the data.…”

Section: Discussionmentioning

confidence: 99%

Using semi-supervised classifiers for credit scoring

Kennedy¹,

Namee²,

Delany³

2013

Journal of the Operational Research Society

View full text Add to dashboard Cite

“…To get reasonably balanced data for classification, we make the target ration for NFP and FP module as recommended to be 65% and 35%, respectively by Khoshgoftaar et al [10]. The DB is done on the selected features only.…”

Section: The Proposed Framework For Software Fault Predictionmentioning

confidence: 99%

“…Hence, they came up with their proposed over-sampling approach, in which the minority class is over-sampled by creating "synthetic" examples rather than by over-sampling with replacement. In conclusion to get reasonably balanced data for classification, we make the target ration for NFP and FP module as recommended to be 65% and 35%, respectively by [10].…”

Section: Data Balancingmentioning

confidence: 99%

“…Nevertheless, software may have thousands of modules and it's very tiresome and time consuming to allocate human as well as financial resources to go through all modules exhaustively. Thus, statistical and Machine Learning (ML) techniques have been employed in most studies so as to predict the fault-prone modules of the software [1] - [8], [10,13], [15] - [19], [23,24], [27,28,33,34]. Therefore, effective prediction of fault-prone software modules can enable to direct test effort and reduce costs, help to manage resources more efficiently, and be useful for software developer.…”

Section: Introductionmentioning

confidence: 99%

See 1 more Smart Citation

A Combined-Learning Based Framework for Improved Software Fault Prediction

Yohannese¹,

Li²

2017

IJCIS

View full text Add to dashboard Cite

Software Fault Prediction (SFP) is found to be vital to predict the fault-proneness of software modules, which allows software engineers to focus development activities on fault-prone modules, thereby prioritize and optimize tests, improve software quality and make better use of resources. In this regard, machine learning has been successfully applied to solve classification problems for SFP. Nevertheless, the presence of different software metrics, the redundant and irrelevant features and the imbalanced nature of software datasets have created more and more challenges for the classification problems. Therefore, the objective of this study is to independently examine software metrics with multiple Feature Selection (FS) combined with Data Balancing (DB) using Synthetic Minority Oversampling Techniques for improving classification performance. Accordingly, a new framework that efficiently handles those challenges in a combined form on both Object Oriented Metrics (OOM) and Static Code Metrics (SCM) datasets is proposed. The experimental results confirm that the prediction performance could be compromised without suitable Feature Selection Techniques (FST). To mitigate that, data must be balanced. Thus our combined technique assures the robust performance. Furthermore, a combination of Random Forts (RF) with Information Gain (IG) FS yields the highest Receiver Operating Characteristic (ROC) curve (0.993) value, which is found to be the best combination when SCM are used, whereas the combination of RF with Correlation-based Feature Selection (CFS) guarantees the highest ROC (0.909) value, which is found to be the best choice when OOM are used. Therefore, as shown in this study, software metrics used to predict the fault proneness of the software modules must be carefully examined and suitable FST for software metrics must be cautiously selected. Moreover, DB must be applied in order to obtain robust performance. In addition to that, dealing with the challenges mentioned above, the proposed framework ensures the remarkable classification performance and lays the pathway to quality assurance of software.

show abstract

Learning with limited minority class data

Cited by 83 publications

References 11 publications

Overlapping, Rare Examples and Class Decomposition in Learning Classifiers from Imbalanced Data

Overlapping, Rare Examples and Class Decomposition in Learning Classifiers from Imbalanced Data

Using semi-supervised classifiers for credit scoring

A Combined-Learning Based Framework for Improved Software Fault Prediction

Contact Info

Product

Resources

About