Skip to main content
. Author manuscript; available in PMC: 2020 Dec 1.
Published in final edited form as: J Occup Environ Med. 2019 Dec;61(Suppl 12):S55–S64. doi: 10.1097/JOM.0000000000001692

Figure 4.

Figure 4.

Features identified by SVM with recursive feature elimination to classify subjects into smokers and non-smokers. Serum biomarkers were analyzed by SVM as described. (A) The smallest feature set to achieve ≥75% training and cross-validation accuracy is shown, with the relative predictive weights. Features of interest are highlighted in red (miRNAs), green (cytokines), blue (cardiovascular markers) and yellow (the tryptophan pathway). Arrows indicate miRNAs selected for further validation. (B) Weighted principal component analysis of the 775 subjects using the 45 feature set. Blue dots represent non-smokers and red dots represent smokers. Using the SVM weights, the PCA is able to classify the subjects by smoking status.