Figure 3.

Comparison of DBSI‐DNN model with varying number of hidden layers and node count. (A) Neural nets with 1 to 20 hidden layers were tested on 70 different random states of our data for validation accuracy. About 80% of the data was used to train the DNN, 10% was used for testing and another 10% for validating. Reliability/predictability of neural networks are modelled by standard deviation. All hidden layers tested contain 100 nodes. (B) The number of epochs required for neural networks to reach 90% validation accuracy are shown. Neural networks with 10 to 200 nodes in each hidden layer were tested. All neural networks contain 10 hidden layers. Each neural network was tested via 10‐fold cross validation tests 10 times. Neural networks with 10 and 20 nodes did not attain a validation accuracy of 90% in any of its trials within 150 epochs, and therefore are not shown in the figure. (C) The optimized neural network of 10 hidden layers each containing 100 nodes was tested on 70 different random states of our data for validation accuracy. This graph shows the validation accuracies over these trials.