Table 3 AUROC values of each model developed using various algorithms and training and testing data sets, and AUROC values of logistic regression after attribute selection using training and testing sets (Italic Data).

Algorithms	Outcomes	No BPD	No BPD or Gr1 BPD	Gr3 BPD or death before 36w PMA	Death before 36w PMA	Overall mortality
kNN	Training data set	0.735	0.720	0.762	0.744	0.746
kNN	Testing data set	0.753	0.708	0.789	0.805	0.792
Logistic regression	Training data set	0.803	0.777	0.811	0.833	0.831
Logistic regression	Testing data set	0.812	0.769	0.854	0.884	0.884
Naïve bayes	Training data set	0.783	0.757	0.789	0.819	0.817
Naïve bayes	Testing data set	0.781	0.736	0.841	0.877	0.879
Neural network	Training data set	0.761	0.735	0.779	0.783	0.785
Neural network	Testing data set	0.766	0.721	0.786	0.814	0.818
Random forest	Training data set	0.765	0.747	0.780	0.784	0.780
Random forest	Testing data set	0.765	0.733	0.819	0.857	0.845
SVM	Training data set	0.645	0.647	0.631	0.659	0.639
SVM	Testing data set	0.664	0.623	0.670	0.804	0.708
Classification tree	Training data set	0.632	0.645	0.583	0.587	0.560
Classification tree	Testing data set	0.682	0.642	0.646	0.608	0.702
After attribute selection
Logistic regression	Training data set	0.802	0.776	0.811	0.835	0.833
Logistic regression	Testing data set	0.801	0.763	0.850	0.881	0.881

AUROC Area under the receiver operating characteristic curve; BPD Bronchopulmonary dysplasia; kNN k-nearest neighbors; Gr Grade; PMA Postmenstrual age; SVM Support vector machine.

Quick links

Search