Skip to main content

Table 1 Traditional Classifiers Used for Feature Set Selection

From: A Bayesian network approach to feature selection in mass spectrometry data

Method

Features Selected

Min Error

Lockbox Error

LDA

7.2

7.8 ± 1.2%

21.0 ± 2.6%

QDA

9.7

3.8 ± 0.5%

38.5 ± 1.4%

Mahalanobis

2.8

11.1 ± 0.5%

28.3 ± 2.8%

NBC

6.8

8.4 ± 0.7%

18.9 ± 1.4%

  1. Four different methods of classification are used to select features from the leukemia data set. 30 trials were performed for each. Features Selected represents the average number of features in a feature set when the wrapper determined no better feature set was available. Min Error represents the average 10-fold stratified cross-validated error rate achieved by the final feature set. Lockbox Error is the average error achieved by using the final feature set (and associated parameters) to classify the data which was withheld until the final classifier was built.