Neural Network-Based Algorithmic Trading Systems: Multi-Timeframe Analysis and High-Frequency Execution in Cryptocurrency Markets
Paper • 2508.02356 • Published • 1
BTC Futures Movement Predictor v4
XGBoost + CNN Ensemble for predicting BTC futures price direction (UP/DOWN).
Architecture
- XGBoost (60% weight): Gradient boosted trees on 79 hand-crafted features
- CNN (40% weight): Conv1D with residual blocks on 63-day windows
- Binary classification: UP (>0.2%) / DOWN (<-0.2%) next-day return
- Wavelet denoising preprocessing (db4, 3 levels)
Data
- BTC/ETH OHLCV: CryptoCompare API (2017-2026, ~3400 days)
- Gold, DXY, VIX: Yahoo Finance
- 79 features: Returns (8 horizons), Volatility (6), MAs (8), EMAs (5), MACD (3), Bollinger (4), RSI (3), ATR (2), Volume (5), Stochastic (4), OBV, Williams %R, MFI, Momentum (4), Calendar (4), Higher-order stats (3), Cross-asset (ETH RSI/correlation/ratio, Gold/DXY/VIX returns & correlations)
Results
| Metric | XGBoost | CNN | Ensemble |
|---|---|---|---|
| Test Accuracy | 0.5396 | 0.5542 | 0.5458 |
| Macro F1 | 0.4572 | 0.4410 | 0.3876 |
| AUC-ROC | 0.4968 | - | 0.5198 |
Backtest
| Metric | Value |
|---|---|
| Strategy Return | -0.13% |
| Buy & Hold Return | -16.19% |
| Sharpe Ratio | -0.895 |
| Max Drawdown | -0.13% |
| Profit Factor | 0.000 |
| Win Rate | 0.0% |
References
- arXiv:2508.02356 (Multi-head CNN + Soft Attention)
- arXiv:2506.22055 (LSTM+XGBoost hybrid)
- arXiv:2408.12408 (Wavelet denoising)
- arXiv:2405.11431 (Conv-LSTM Multivariate)
⚠️ Disclaimer
Research only. Not financial advice. Crypto markets are extremely volatile.
Inference Providers NEW
This model isn't deployed by any Inference Provider. 🙋 Ask for provider support
Papers for SummerOfHaze/btc-futures-predictor
crypto price prediction using lstm+xgboost
Paper • 2506.22055 • Published
An Evaluation of Deep Learning Models for Stock Market Trend Prediction
Paper • 2408.12408 • Published • 1
Review of deep learning models for crypto price prediction: implementation and evaluation
Paper • 2405.11431 • Published