Table 5 Hyperparameters used to tune the DNN model, their range, and the result of tuning.
Hyperparameter | Description | Range | Selected |
|---|---|---|---|
Layers | Individual building blocks of a neural network | Range (2, 5) | 4 |
Neurons | Basic computational units within a neural network | Arange (2, 256, 4) | 64, 256, 64, 1 |
Dropout | Regularization technique commonly used in neural networks to prevent overfitting | Arange (0.20, 0.75, 0.025) | 0.7 |
Batch size | How many training examples are processed in a single forward/backward pass during each training iteration | Arange (4, 128, 8) | 44 |
Activation function | Neural networks can learn intricate relationships between inputs and outputs by adding non-linearities to a neuron's output | Relu, tanh, sigmoid | Sigmoid |
Optimizer | Algorithm used to modify a neural network's weights and biases while it is being trained | Adadelta, adam, rmsprop, adagrad, sgd | Adam |
