Fig 1.
iDeepLe flowchart.
Table 1.
Parameters setting for the grid-search stage 1.
Fig 2.
The architecture of the optimized deep neural network model.
Table 2.
Properties of the activation functions used in different layers of iDeepLe.
Table 3.
Parameters setting for the grid-search stage 2.
Table 4.
Optimized hyper-parameters for the optimization algorithms after the grid-search stage 2.
Fig 3.
Density plots of the predictor input variables.
Fig 4.
Scatter matrix presentation of the predictor input variables with their probability histograms.
Table 5.
Parameters setting for the grid-search stage 3.
Fig 5.
10-folds cross-validation R2 during number of epochs for different batch sizes employing different optimization algorithms.
Fig 6.
Statistical metrics for different batch sizes employing different optimization algorithms.
Fig 7.
Histograms of the ratio of the predicted and the measured Ln(SA) for different batch sizes.
Mean and coefficient of variation of this ratio are also reported for different batch sizes. CB 2008 model has also been demonstrated for comparison.
Table 6.
External validation results of the deep models with different batch sizes and CB 2008 model.
Table 7.
Parameters setting for the Adam optimizer using the popular deep learning libraries.
Fig 8.
Radar plots of the regression metrics for the most common regression models compared to iDeepLe model.