|
|
--- |
|
|
license: apache-2.0 |
|
|
--- |
|
|
|
|
|
# MLCryptoForecasterICH.py |
|
|
|
|
|
H4 model is performing quite well: |
|
|
|
|
|
- **Neutral (inside cloud, –1)**: |
|
|
- Precision 0.69 — when it predicts “neutral,” it’s correct 69% of the time. |
|
|
- Recall 0.43 — it only catches 43% of the actual neutral periods (you might tighten this). |
|
|
|
|
|
- **Downtrend (0)**: |
|
|
- Precision 0.81, Recall 0.93 — it’s very good at spotting bearish moves. |
|
|
|
|
|
- **Uptrend (1)**: |
|
|
- Precision 0.90, Recall 0.94 — excellent at capturing rallies. |
|
|
|
|
|
- **Overall accuracy: 84%** — a big jump from ~50% on 15 min data. |
|
|
|
|
|
**Ichimoku cloud features** (plus the suite of other indicators) really helped the model understand trend context. |
|
|
|
|
|
--- |
|
|
|
|
|
### Next steps you might consider: |
|
|
|
|
|
1. **Feature Importance** |
|
|
Plot a bar chart of your model’s feature importances to see which indicators are driving predictions. |
|
|
|
|
|
2. **Backtesting** |
|
|
Build a simple backtester that uses your predicted signals to simulate entry/exit and evaluate net returns. |
|
|
|
|
|
3. **Hyperparameter Tuning** |
|
|
Run a grid search or randomized search to squeeze out even more performance. |
|
|
|
|
|
4. **Visualization** |
|
|
Overlay your up/down/neutral predictions on a price+cloud chart for visual validation. |
|
|
|
|
|
--- |
|
|
|
|
|
### In all of our logging and modeling: |
|
|
|
|
|
1 = Uptrend |
|
|
|
|
|
0 = Downtrend |
|
|
|
|
|
–1 = Neutral (price inside the Ichimoku cloud) |
|
|
|
