A Machine-Learned Predictor of Colonic Polyps Based on Urinary Metabolomics

Table 5

Feature selection methods used to select 10 metabolites, while using a LASSO classifier and all 4 clinical features.

Method	Average AUC	AUC standard; deviation
Random	0.678	0.036
Correlation (Pearson)	0.712	0.008
mRMR	0.707	0.018
Mutual information	0.712	0.015
SVM weights	0.699	0.020
SVM recursive	0.697	0.016

The bolded row shows the approach we have decided to use.