Skip to main content

Table 4 Percentage error rates for the Bayesian Model (BM), GBM and Random Forest (RF) after filtering calls on the percentage of a serotype’s cps genes significantly present, using a 50% threshold; for samples containing mixtures, samples containing single serotypes and all samples combined

From: A comparison of machine learning and Bayesian modelling for molecular serotyping

Dataset

BM

GBM

RF

D.36

Mixtures

15

(12.4-17.8)

5.9

(4.3-7.8)

9.1

(7.1-11.3)

 

Singles

4.8

(4.1-5.5)

1.5

(1.1-1.9)

3.3

(2.7-3.9)

 

Combined

6.5

(5.8-7.3)

2.2

(1.8-2.7)

4.3

(3.7-4.9)

D.73

Mixtures

19

(16.5-21.6)

12

(10.0-14.2)

16

(13.7-18.4)

 

Singles

5.6

(4.9-6.4)

5.0

(4.3-5.7)

7.1

(6.3-7.9)

 

Combined

8.3

(7.5-9.1)

6.4

(5.7-7.1)

8.9

(8.1-9.7)

  1. Figures in brackets show 95% credible interval