Table 4 Hyperparameters for the best models for predicting magnitude of change Acc. F1 Score - The sum of the individual F1 Scores of each label/class
From: Bitcoin price change and trend prediction through twitter sentiment and data volume
LSTM | CNN | BiLSTM | |
|---|---|---|---|
# Layers | 3 | 2 | 2 |
Layer Size | 256 | 128 | 256 |
Batch Size | 50 | 80 | 80 |
Dataset | 1 day lag | 3 day lag | 1 day lag |
Lagged Features | 7 | 3 | 7 |
Train-Test Split | 85:15 | 85:15 | 85:15 |
Loss Function | Categorical Crossentropy | Categorical Crossentropy | Categorical Crossentropy |
Early Stopping Parameter | Validation Loss | Validation Loss | Validation Loss |
Early Stopping Patience | 20 | 20 | 20 |
Maximum Accuracy | 58.21% | 57.35% | 59.09% |
Mean Accuracy | 46.76% | 51.47% | 46.67% |
Acc. F1 Score | 12.33% | 14.21% | 12.88% |