A Data Mining Approach Identified Salivary Biomarkers That Discriminate between Two Obesity Measures

Table 1

Important variables associated with OBW and OBWHO as identified by three algorithms.

Methods for the two obesity measures	Top-ranked variables (in the order of relative importance)	AUC of respective classification models
Obesity (OBW)
Lasso	Insulin, CRP, fitness, adiponectin, leptin	0.787
MARS	CRP, insulin, adiponectin, fitness, VEGF, leptin	0.837
Random forest	CRP, insulin, leptin	0.826
Boosting classification trees	CRP, insulin, adiponectin, leptin, fitness	0.816
Obesity (OBWHO)
Lasso	Insulin, CRP, adiponectin, fitness	0.816
MARS	CRP, insulin, adiponectin, sex, VEGF, fitness	0.853
Random forest	CRP, insulin, adiponectin	0.833
Boosting classification trees	CRP, insulin, adiponectin	0.822

Variables with relative importance scores ≥ 20%.