Predicting asthma control deterioration in children

Table 3 Performance of the different classifiers

Performance of the decision stump classifier
Evaluation method	Sensitivity	Accuracy	Specificity	AUC	PPV^a	NPV^a
10-fold cross validation	67.2 %	73.4 %	74.9 %	0.710	38.1 %	91.0 %
testing on each patient’s last assessment	51.1 %	73.9 %	82.0 %	0.665	50.0 %	82.6 %
Performance of the six advanced classifiers measured by the 10-fold cross validation method
Classifier	Sensitivity	Accuracy	Specificity	AUC	PPV^a	NPV^a
Multiboost with decision stumps	73.8 %	71.8 %	71.4 %	0.761	37.1 %	92.4 %
Support vector machine	71.5 %	72.0 %	72.0 %	0.718	37.0 %	91.8 %
Deep learning	71.6 %	72.3 %	72.5 %	0.744	37.2 %	91.8 %
Naive Bayes	59.8 %	78.1 %	82.3 %	0.777	43.7 %	90.0 %
k-nearest neighbor	56.9 %	73.3 %	77.0 %	0.704	36.0 %	88.7 %
Random forest	48.3 %	75.8 %	82.0 %	0.662	37.9 %	87.5 %
Performance of the six advanced classifiers measured by the method of testing on each patient’s last assessment
Classifier	Sensitivity	Accuracy	Specificity	AUC	PPV^a	NPV^a
Multiboost with decision stumps	74.5 %	74.4 %	74.4 %	0.757	50.7 %	89.2 %
Support vector machine	70.2 %	73.3 %	74.4 %	0.723	49.3 %	87.6 %
Deep learning	68.1 %	72.2 %	73.7 %	0.738	47.8 %	86.7 %
Naive Bayes	44.7 %	73.9 %	84.2 %	0.783	50.0 %	81.2 %
k-nearest neighbor	48.9 %	73.9 %	82.7 %	0.773	50.0 %	82.1 %
Random forest	38.3 %	75.6 %	88.7 %	0.678	54.5 %	80.3 %

ISSN: 1472-6947