Skip to main content

Table 2 Best performing LSTM network configurations for FOG detection

From: Prediction and detection of freezing of gait in Parkinson’s disease from plantar pressure data using long short-term memory neural-networks

Network or training parameter Values / Options
LSTM layers (units in each LSTM layer) 2 layers (16 units) and 3 layers (32 units)
Initial learning rate  0.01
Learning rate decay Decreases to half, every 5 epochs
Optimizer Adam optimizer
Loss function Cross entropy loss function
Batch size 1
Training epochs 30