|
|
---
|
|
|
license: mit
|
|
|
tags:
|
|
|
- finance
|
|
|
- trading
|
|
|
- bitcoin
|
|
|
- cryptocurrency
|
|
|
- machine-learning
|
|
|
- deep-learning
|
|
|
- lstm
|
|
|
- transformer
|
|
|
- xgboost
|
|
|
- random-forest
|
|
|
- shap
|
|
|
language:
|
|
|
- en
|
|
|
library_name: pytorch
|
|
|
pipeline_tag: tabular-classification
|
|
|
---
|
|
|
|
|
|
# 📈 AI Multi-Model Trading Bot
|
|
|
|
|
|
A comprehensive cryptocurrency trading signal prediction system using 8 ML/DL models with ensemble voting.
|
|
|
|
|
|
## 🎯 Model Overview
|
|
|
|
|
|
| Model Type | Models Included |
|
|
|
| :--- | :--- |
|
|
|
| **Traditional ML** | Logistic Regression, Random Forest, XGBoost |
|
|
|
| **Deep Learning** | LSTM, GRU, CNN, LSTM+Attention, Transformer |
|
|
|
| **Ensemble** | Majority voting across all models |
|
|
|
|
|
|
## 📊 Features Used
|
|
|
|
|
|
The models use 10 technical indicators:
|
|
|
- RSI (Relative Strength Index)
|
|
|
- MACD & MACD Signal
|
|
|
- Bollinger Band Width
|
|
|
- ATR (Average True Range)
|
|
|
- Distance from SMA50
|
|
|
- OBV Percentage Change
|
|
|
- ADX (Average Directional Index)
|
|
|
- Stochastic RSI (K & D)
|
|
|
|
|
|
## 🚀 Quick Start
|
|
|
|
|
|
```python
|
|
|
import joblib
|
|
|
import torch
|
|
|
|
|
|
# Load scaler and config
|
|
|
scaler = joblib.load("scaler.pkl")
|
|
|
config = joblib.load("config.pkl")
|
|
|
|
|
|
# Load ML model
|
|
|
rf_model = joblib.load("random_forest.pkl")
|
|
|
|
|
|
# Load DL model
|
|
|
from your_models import LSTMModel
|
|
|
lstm = LSTMModel(config['input_dim'])
|
|
|
lstm.load_state_dict(torch.load("lstm.pt"))
|
|
|
lstm.eval()
|
|
|
```
|
|
|
|
|
|
## 📁 Files
|
|
|
|
|
|
| File | Description |
|
|
|
| :--- | :--- |
|
|
|
| `scaler.pkl` | StandardScaler for feature preprocessing |
|
|
|
| `config.pkl` | Model configuration (input_dim, timesteps, feature_cols) |
|
|
|
| `logistic_regression.pkl` | Trained Logistic Regression model |
|
|
|
| `random_forest.pkl` | Trained Random Forest model |
|
|
|
| `xgboost.pkl` | Trained XGBoost model |
|
|
|
| `lstm.pt` | Trained LSTM model weights |
|
|
|
| `gru.pt` | Trained GRU model weights |
|
|
|
| `cnn.pt` | Trained CNN model weights |
|
|
|
| `lstm_attention.pt` | Trained LSTM+Attention model weights |
|
|
|
| `transformer.pt` | Trained Transformer model weights |
|
|
|
| `shap_values.pkl` | SHAP feature importance values |
|
|
|
|
|
|
## 📊 Dataset
|
|
|
|
|
|
Training data is available separately:
|
|
|
**🔗 [AdityaaXD/Multi-Model-Trading-Data](https://huggingface.co/datasets/AdityaaXD/Multi-Model-Trading-Data)**
|
|
|
|
|
|
- **Ticker**: BTC-USD
|
|
|
- **Date Range**: 2015-01-01 to 2025-01-01
|
|
|
- **Total Samples**: ~3,600 days
|
|
|
- **Train/Test Split**: 80/20
|
|
|
|
|
|
## ⚠️ Disclaimer
|
|
|
|
|
|
This model is for **educational and research purposes only**. It should NOT be used for actual trading decisions. Cryptocurrency markets are highly volatile and past performance does not guarantee future results.
|
|
|
|
|
|
## 📊 SHAP Explainability
|
|
|
|
|
|
The model includes SHAP (SHapley Additive exPlanations) values for feature importance analysis, helping understand which technical indicators most influence predictions.
|
|
|
|
|
|
## 🛠️ Training Details
|
|
|
|
|
|
- **Hyperparameter Tuning**: GridSearchCV with 3-fold CV
|
|
|
- **Deep Learning**: 50 epochs, early stopping (patience=7)
|
|
|
- **Regularization**: Label smoothing (0.1), gradient clipping (1.0)
|
|
|
- **Class Balancing**: Weighted loss functions
|
|
|
|
|
|
## 📝 Citation
|
|
|
|
|
|
```bibtex
|
|
|
@misc{ai-trading-bot-2025,
|
|
|
title={AI Multi-Model Trading Bot},
|
|
|
author={Your Name},
|
|
|
year={2025},
|
|
|
publisher={Hugging Face}
|
|
|
}
|
|
|
```
|
|
|
|
