Irrelevant Features and the Subset Selection Problem

By George H. John, Ron Kohavi, Karl Pfleger. 1994

How researchers see this paper

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

Citations in the corpus

Relational autoencoder for feature extraction (2017, methodology): “Although feature selection methods, such as Subset Selection [5] and Random Forest [6], are effective in filtering out unnecessary features, if all input features contribute to final results to some extent, feature selection is more likely to give rise to information loss than feature extraction bec”
Efficient Feature Selection via Analysis of Relevance and Redundancy (2004, methodology): “The ﬁrst data set is the widely used Corral data (John et al., 1994) which contains six Boolean features ( A 0, A 1, B 0, B 1, I , R ) and a Boolean class Y deﬁned by Y = ( A 0 ∧ A 1 ) ∨ ( B 0 ∧ B 1 ) .”
Feature Generation Using General Constructor Functions (2002, methodology): “Methods similar to those employed for feature selection ( John, Kohavi, & Pfleger, 1994; Kohavi & Dan, 1995; Salzberg, 1993) could be adopted to find a utile subset of constructor functions.”
Filters, Wrappers and a Boosting-Based Hybrid for Feature Selection (2001, methodology): “Wrapper methods, e.g (John et al., 1994; Langley & Sage, 1994; Caruana & Freitag, 1994, inter alia), search through the space of feature subsets using a learning algorithm to inform the search.”
Redundancy based feature selection for microarray data (2004, background): “Based on a review of previous definitions of feature relevance, John, Kohavi, and Pfleger classified features into three disjoint categories, namely, strongly relevant, weakly relevant, and irrelevant features [11].”
Linkage and Autocorrelation Cause Feature Selection Bias in Relational Learning (2002, background): “This definition is broader than some (e.g., John, Kohavi, and Pfleger 1994), but it emphasizes the central role of feature selection in machine learning algorithms, including algorithms for learning decision trees, classification and association rules, and Bayesian nets.”
Evolutionary joint selection to improve human action recognition with RGB-D devices (2014, methodology): “It uses it in order to perform evaluations on the possible feature subset selections and to find the optimal one (John et al., 1994).”
Survey on Feature Selection (2015, methodology): “Wrapper methods use the learning algorithm, during the evaluation step, to determine the utility of a certain features subset based on the algorithm’s accuracy while using that speciﬁc subset (John et al., 1994).”
Multi-scale digital terrain analysis and feature selection for digital soil mapping (2010, methodology): “The importance of a feature, or the predictive power of a feature, can be calculated in two ways – based on so-called filter or wrapper approaches (John et al., 1994).”
Dimensionality reduction based on rough set theory: A review (2009, background): “Feature selection has been studied intensively in the past one decade [4,49,52,68].”
Feature selection algorithms: a survey and experimental evaluation (2002, methodology): “Wrapper Scheme: In this scheme the relationship is taken the other way around: it is the FSA that uses the learning algorithm as a subroutine[ 11 ].”
Legal requirements on explainability in machine learning (2020, background): “Some features can also be characterized as weakly relevant if they bring new information, but only if other features are removed (John et al. 1994; Kohavi and John 1997; Frénay et al. 2014).”