Skip to main content

Table 1 Summary of predictive accuracy of random forest supervised learning models

From: Forensic analysis of the microbiome of phones and shoes

Sample subset

Predicted category

N

Estimated error ± SD

Baseline error

Ratio

All phone samples

Person

104

0.037 ± 0.062

0.500

13.63

All shoe samples

Person

211

0.010 ± 0.020

0.479

50.26

P1 phone samples

Front/back

52

0.417 ± 0.206

0.481

1.15

P2 phone samples

Front/back

52

0.268 ± 0.180

0.481

1.79

P1 shoe samples

Shoe surface

110

0.705 ± 0.125

0.736

1.05

P2 shoe samples

Shoe surface

101

0.796 ± 0.090

0.732

0.92

  1. Tenfold cross validation models were constructed with 1,000 trees using OTUs from evenly rarified samples as predictors of sample origin. P1, person 1; P2, person 2, SD, standard deviation, N, number.