Machine learning methods to predict child posttraumatic stress: a proof of concept study

BMC Psychiatry

Table 1 Performance of classifiers and feature selection methods

Classifier		All features	Feature selection with HITON-PC
SVM linear	observed data	0.79 (0.02)**	0.68 (0.04)*
SVM linear	label shuffling	0.50 [0.32 0.71]	0.50 [0.36 0.67]
SVM poly	observed data	0.78 (0.02)*	0.68 (0.04)
SVM poly	label shuffling	0.50 [0.31 0.71]	0.50 [0.34 0.71]
SVM RB	observed data	0.76 (0.02)*	0.68 (0.04)
SVM RB	label shuffling	0.50 [0.36 0.70]	0.50 [0.35 0.69]
Random forest	observed data	0.78 (0.01)**	0.74 (0.01)*
Random forest	label shuffling	0.50 [0.33 0.67]	0.50 [0.33 0.73]
Lasso	observed data	0.67 (0.01)**	0.74 (0.01)*
Lasso	label shuffling	0.50 [0.44 0.57]	0.50 [0.35 0.68]
Logistic Regression (LR)	observed data	0.47 (0.01)	0.72 (0.01)
Logistic Regression (LR)	label shuffling	0.50 [0.35 0.64]	0.51 [0.32 0.74]
Stepwise LR	observed data	0.57 (0.02)	0.72 (0.02)*
Stepwise LR	label shuffling	0.51 [0.39 0.64]	0.49 [0.31 0.71]

The performance (measured as Area Under the ROC Curve) of individual classifiers and feature selection methods in the observed data and under the null hypothesis of no signal in the data (estimated with label shuffling). For observed results the mean and (standard deviation) were presented. For the label shuffling, mean and [95% confidence interval] were presented. Predictive performance that are significant at p < 0.05 is labeled with *, predictive performance that are significant at p < 0.01 is labeled with **

ISSN: 1471-244X