Hidden Layer 1

The number of neurons in the first processing layer. More neurons = the model can learn more complex patterns, but risks memorizing the training data instead of generalizing. → Leave at 64.

Hidden Layer 2

An optional second processing layer for deeper pattern recognition. Set to None for a simpler 2-layer network. Adding a second layer helps with complex market patterns but increases training time. → Leave at 32.

Activation Function

Determines how each neuron processes its input. ReLU is the standard choice — fast and effective for most cases. Tanh and ELU are alternatives with slightly different behavior on edge cases. → Leave at ReLU.

Dropout 0.30

During training, randomly ignores this percentage of neurons. This forces the network to learn robust patterns rather than memorizing the data (overfitting). 0.30 means 30% are randomly disabled each step. → Leave at 0.30.

Learning Rate

Controls how much the model adjusts itself after each mistake. Too high = learns fast but imprecisely. Too low = very precise but slow. 0.001 is the well-tested default for most financial time series. → Leave at 0.001.

Max Epochs 100

One epoch = one full pass through all training data. The model trains up to this limit, but Early Stopping will automatically finish training earlier if the model stops improving — so the actual number is usually lower. → Leave at 100.