Table 3

Supervised learning classification using three different algorithms: k-nearest neighbors, support vector machine, and naïve Bayes classificationa

Age


Population

6 months

9 months

12 months

18 months

24 months


k-NN

0.67

(0.06)

0.77

(0.02)

0.53

(0.38)

0.72

(0.12)

0.53

(0.47)

All infants Accuracy (P value)

SVM

0.63

(0.16)

0.77

(0.00)

0.53

(0.71)

0.65

(0.56)

0.55

(0.64)

Bayes

0.70

(0.05)

0.72

(0.03)

0.68

(0.06)

0.80

(0.04)

0.57

(0.33)

k-NN

0.40

(0.64)

0.90

(0.00)

0.70

(0.16)

0.90

(0.03)

-

Boys Accuracy (P value)

SVM

0.30

(0.42)

1.00

(0.00)

0.75

(0.12)

0.75

(0.81)

-

Bayes

0.35

(0.58)

0.75

(0.10)

0.75

(0.09)

0.90

(0.05)

-

k-NN

0.80

(0.03)

0.60

(0.20)

0.48

(0.58)

0.35

(0.88)

0.40

(0.89)

Girls Accuracy (P value)

SVM

0.80

(0.02)

0.40

(0.54)

0.35

(0.97)

0.55

(0.78)

0.75

(0.53)

Bayes

0.75

(0.07)

0.65

(0.19)

0.47

(0.54)

0.45

(0.73)

0.50

(0.92)


aTenfold cross-validation was run using the computed mean mMSE values on 64 channels for each infant within each age group. P values were estimated empirically using a permutation of class labels approach as described in the methods section under 'classification and endophenotypes. Identical cross-validation calculations with 100 permutations were performed to determine empirical P values with three different populations: all infants, boys only and girls only. Too few 24-month-old boys were available for cross-validation. k-NN, k-nearest neighbors algorithm; SVM, support vector machine algorithm; Bayes, naïve Bayes classification algorithm. Boldface entries highlight values with statistical significance of p < 0.05.

Bosl et al. BMC Medicine 2011 9:18   doi:10.1186/1741-7015-9-18

Open Data