ML Model Fleet
Planned ML model fleet — backtesting metrics from research phase
ML Pipeline — Development Roadmap
These models are planned capabilities. Metrics shown are from backtesting research, not live trading. The ensemble system will integrate with Trinity consensus once deployed.
Model Fleet Overview
Random Forest
Accuracy
62.3%
Gradient Boosting (XGBoost)
Accuracy
65.7%
Ridge Regression
Accuracy
58.1%
LSTM Neural Network
Accuracy
64.1%
Support Vector Regression
Accuracy
59.2%
Performance Comparison
| Model | Accuracy | Precision | Recall | F1 Score | AUC-ROC | Sharpe Ratio |
|---|---|---|---|---|---|---|
RF | 62.3% | 64.1% | 59.8% | 61.9% | 67.1% | 1.42 |
XGB | 65.7% | 67.2% | 63.1% | 65.1% | 70.9% | 1.67 |
RIDGE | 58.1% | 59.4% | 56.7% | 58.0% | 62.3% | 1.11 |
LSTM | 64.1% | 65.8% | 61.2% | 63.4% | 69.4% | 1.53 |
SVR | 59.2% | 60.8% | 57.4% | 59.1% | 63.8% | 1.19 |
Feature Engineering
| Feature | Category | RF | XGB | RIDGE | LSTM | SVR |
|---|---|---|---|---|---|---|
| Price Momentum | Technical | + | + | + | + | + |
| Volume Profile | Technical | + | + | + | + | + |
| Sentiment Score | Alternative | + | + | - | + | + |
| Options Flow | Derivatives | + | + | - | + | - |
| Insider Activity | Fundamental | + | + | - | - | + |
| Macro Indicators | Macro | + | + | + | + | + |
| Sector Rotation | Macro | + | + | + | - | - |
| Earnings Surprise History | Fundamental | + | + | + | + | - |
| Total Features | 8/8 | 8/8 | 5/8 | 6/8 | 5/8 | |
Backtesting Results
| Model | Total Trades | Win Rate | Avg Return/Trade | Max Drawdown | Profit Factor | Calmar Ratio |
|---|---|---|---|---|---|---|
| RF | 541 | 60.1% | +0.41% | -6.8% | 1.38 | 1.22 |
| XGB | 541 | 64.3% | +0.47% | -5.9% | 1.52 | 1.41 |
| RIDGE | 541 | 56.8% | +0.27% | -7.4% | 1.22 | 0.89 |
| LSTM | 468 | 62.2% | +0.43% | -6.4% | 1.44 | 1.28 |
| SVR | 413 | 57.6% | +0.29% | -8.2% | 1.25 | 0.91 |
Model Ensemble
Weighted average ensemble — final signal is a confidence-weighted combination of all active model predictions.
Signal Calculation
S_final = sum(w_i * p_i) / sum(w_i)
where w_i = model weight, p_i = model prediction
Training Pipeline
Data Collection
Completed Feb 14, 8:00 AM
Feature Engineering
Completed Feb 14, 8:04 AM
Model Training
In progress...
Validation
Waiting
Deployment
Waiting