A distinct metabolic signature predicts development of fasting plasma glucose

Table 4 Regression model accuracy

Metabolite Selection	Model	all samples	CV
Spearman Correlation	linear Model	0.57	0.22
Spearman Correlation	Random Forest	0.97	0.41
RF importance	Random Forest	0.97	0.47
RF importance + Established markers	Random Forest	0.97	0.46
Established markers	Random Forest	0.90	0.05

Accuracy of the models (median Pearson correlation between real and estimated Δglucose levels) based on metabolites and/or established risk markers was calculated using all samples of the training set and after tenfold cross-validation (CV). The established risk markers are: gender, waist circumference, BMI, age and baseline fasting glucose levels.