Fig 1.
General architecture of the proposed AE-DNNs method.
DAE, deep autoencoder; DNN, deep neural network; RE, reconstruction error; CHD, coronary heart disease.
Fig 2.
A simple autoencoder neural network architecture based on a fully-connected layer.
Fig 3.
A simple neural network architecture with one hidden layer.
Fig 4.
Design of experimental study for the AE-DNNs.
Steps 1-7 (green line) represent model training, and steps 8-13 (blue line) show model evaluation. KNHANES, Korea national health and nutrition examination survey; AE, autoencoder; RE, reconstruction error; DNN, deep neural network; CHD, coronary heart disease.
Fig 5.
Integrated KNHANES dataset.
Table 1.
Description of the selected features.
Table 2.
Result of feature ranking.
Table 3.
Results of DNN models with recursive feature elimination.
Fig 6.
Comparison of the selected features and other guidelines.
The feature group 1 is the Framingham risk factors, and the feature group 2 was proposed by [18-20] papers on KNHANES dataset.
Table 4.
Results of compared algorithms on integrated KNHANES dataset.
Table 5.
Results of the receiver operating characteristic curve analysis of compared algorithms.
Fig 7.
Receiver operating characteristic curves of compared algorithms.
(a) Receiver operating characteristic curves of Naïve Bayes algorithm; (b) Receiver operating characteristic curves of k-nearest neighbors algorithm; (c) Receiver operating characteristic curves of decision tree algorithm; (d) Receiver operating characteristic curves of support vector machine algorithm; (e) Receiver operating characteristic curves of random forest algorithm; (f) Receiver operating characteristic curves of principal component analysis based deep neural networks (PCA-DNNs) algorithm; (g) Receiver operating characteristic curves of autoencoder based deep neural networks (AE-DNNs) algorithm.
Fig 8.
The average receiver operating characteristic curves of compared algorithms.
Fig 9.
Improvement of area under the curve scores based on input features and the number of deep neural network classifiers.