# ETHUSDT Adaptive threshold classifier 15m

This strategy implements an adaptive threshold classifier approach for ETHUSDT on 15-minute candles using LightGBM.

## Overview

- **Pair**: ETHUSDT
- **Timeframe**: 15m
- **Model**: LightGBM with feature engineering focused on adaptive threshold classification
- **Goal**: Dynamically adjust classification thresholds based on recent volatility and market conditions

## Features Engineered

1. **Basic returns**: 1, 2, 4, 8, 16, 32 period returns
2. **ATR-based features** (enhanced from original):
   - ATR percentage (ATR/price)
   - ATR expansion ratio (short-term/long-term ATR)
   - Realized volatility at different timeframes
   - Volatility regime indicator
   - Volume z-score
   - EMA crossovers (9/34 and 21/89)
3. **Adaptive threshold features**:
   - Dynamic threshold (2× ATR percentage)
   - Volatility percentile ranking
4. **Momentum features**:
   - RSI (Relative Strength Index)
   - Price position in recent 20-period range
   - Volume-price correlation
5. **Distance from EMAs** (from idea): 20, 50, and 200 period EMAs
6. **Prior swing high and swing low distance** (from idea)
7. **Volume features**: Z-score and ratio to moving average
8. **Rolling volatility percentile** (from idea): Fast/slow volatility ratio, volatility percentile

## Configuration

See `quant.config.json` for hyperparameters:
- `lookback`: 100 candles for prediction
- `horizon`: 3 candles forward for labeling (45 minutes for 15m timeframe)
- `threshold`: 0.003 (base threshold, adapted dynamically)
- `min_confidence`: 0.50 minimum probability for signal generation

## Usage

This template follows the PyP Quant Mode contract:
```python
def train(data, config):
    return model, metrics

def predict(model, market_data, config):
    return {"signal": "UP|DOWN|HOLD", "confidence": 0.0, "metadata": {}}
```

## Disclaimer

Educational template only. Not financial advice. Past performance does not guarantee future results.