A review of the stability of feature selection techniques for bioinformatics data

Awada, Wael; Khoshgoftaar, Taghi M.; Dittman, David J.; Wald, Randall; Napolitano, Amri

doi:10.1109/iri.2012.6303031

Cited by 86 publications

(63 citation statements)

References 32 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…People only need to determine the number of model candidate selected in each layer, which is very easy task for nearly all the problems. Third, one advantage of ensemble feature selection [1] is stability. Previous research has shown that feature selection is not stable on imbalanced datasets [42] and so we adopt ensemble of feature selection to make the results more stable.…”

Section: Transferred Feature Selection Based On Gmdh (Tfsg)mentioning

confidence: 99%

A new transferred feature selection algorithm for customer identification

Zhu¹,

Niu²,

Xiao³

et al. 2016

Neural Comput & Applic

View full text Add to dashboard Cite

Class imbalance brings great challenges to feature selection in customer identification and most of current feature selection approaches cannot produce good prediction on the minority class. A number of studies have attempted to solve this issue by using resampling techniques. However, resampling techniques only use the in-domain information and they cannot achieve good performance when the imbalance is caused by the absolute rarity of the minority class. In this paper, we focus on the issue of feature selection with class imbalance caused by absolute rarity. By introducing the idea of transfer learning, we develop a transferred feature selection method based on the group method of data handling neural network (GMDH). The proposed ensemble neural network extracts information of similar customers from related domains to deal with the information scarcity of the minority class in the target domain. Experiments are done on a real-world application from cigarette company. The results indicate that the new method gives better predictive performance than other benchmark feature selection methods, especially the predictive accuracy of minority high-value customers. At the same time, the new algorithm can help to identify important features that distinguish highvalue customers from low-value ones.

show abstract

Section: Transferred Feature Selection Based On Gmdh (Tfsg)mentioning

confidence: 99%

A new transferred feature selection algorithm for customer identification

Zhu¹,

Niu²,

Xiao³

et al. 2016

Neural Comput & Applic

View full text Add to dashboard Cite

show abstract

“…One possible cause is that majority of FS algorithms are designed without consideration of stability aspects, and aim for selecting a minimal subset of the features with the highest classification accuracy [21], [44]. Another cause is existence of the multiple sets of true markers, i.e.…”

Section: Feature Selection Stabilitymentioning

confidence: 99%

“…It was shown that some stability measures are not suitable for large datasets and variable subset size [11], [20].This and other aspect of feature selection are described in several theoretical surveys such as [21], [22], [23]. We used two established and robust stability measures kuncheva index and consistency index.…”

Section: Introductionmentioning

confidence: 99%

An experimental comparison of feature selection methods on two-class biomedical datasets

Drotár

Gazda

Smékal

2015

Computers in Biology and Medicine

View full text Add to dashboard Cite

“…An unstable FS method is generally believed to having little value [21]. As a consequence, the confidence level in selecting optimal features would surely get reduced due to the instability of feature selection results [22].…”

Section: B Stability Evaluationmentioning

confidence: 99%

Weighted heuristic ensemble of filters

Aldehim

Wang

2015

2015 SAI Intelligent Systems Conference (IntelliSys)

View full text Add to dashboard Cite

Abstract-Feature selection has become increasingly important in data mining in recent years due to the rapid increase in the dimensionality of big data. However, the reliability and consistency of feature selection methods (filters) vary considerably on different data and no single filter performs consistently well under various conditions. Therefore, feature selection ensemble has been investigated recently to provide more reliable and effective results than any individual one but all the existing feature selection ensemble treat the feature selection methods equally regardless of their performance. In this paper, we present a novel framework which applies weighted feature selection ensemble through proposing a systemic way of adding different weights to the feature selection methods-filters. Also, we investigate how to determine the appropriate weight for each filter in an ensemble. Experiments based on ten benchmark datasets show that theoretically and intuitively adding more weight to 'good filters' should lead to better results but in reality it is very uncertain. This assumption was found to be correct for some examples in our experiment. However, for other situations, filters which had been assumed to perform well showed bad performance leading to even worse results. Therefore adding weight to filters might not achieve much in accuracy terms, in addition to increasing complexity, time consumption and clearly decreasing the stability.

show abstract

A review of the stability of feature selection techniques for bioinformatics data

Cited by 86 publications

References 32 publications

A new transferred feature selection algorithm for customer identification

A new transferred feature selection algorithm for customer identification

An experimental comparison of feature selection methods on two-class biomedical datasets

Weighted heuristic ensemble of filters

Contact Info

Product

Resources

About