https://github.com/burhanahmed1/cryptosynth

Bitcoin Sentiment Forecast is a Multimodal approach to Bitcoin price forecasting using NLP and Time Series Analysis
https://github.com/burhanahmed1/cryptosynth

datafusion datapreprocessing eda explainable-ai featureengineering machinelearning multimodal-deep-learning pca predictive-modeling

Last synced: 3 months ago
JSON representation

Bitcoin Sentiment Forecast is a Multimodal approach to Bitcoin price forecasting using NLP and Time Series Analysis

• Host: GitHub
• URL: https://github.com/burhanahmed1/cryptosynth
• Owner: burhanahmed1
• License: apache-2.0
• Created: 2025-03-15T10:55:29.000Z (11 months ago)
• Default Branch: main
• Last Pushed: 2025-05-17T05:04:41.000Z (8 months ago)
• Last Synced: 2025-06-14T19:35:19.087Z (8 months ago)
• Topics: datafusion, datapreprocessing, eda, explainable-ai, featureengineering, machinelearning, multimodal-deep-learning, pca, predictive-modeling
• Language: Jupyter Notebook
• Homepage:
• Size: 3.54 MB
• Stars: 0
• Watchers: 1
• Forks: 0
• Open Issues: 0
• Metadata Files:
  • Readme: README.md
  • License: LICENSE

Awesome Lists containing this project

README

          
# CryptoSynth: A Multimodal Generative Framework for Bitcoin Price Forecasting

![Bitcoin Pipeline](architecture/bird-eye.png)

A comprehensive pipeline integrating sentiment analysis of social media data with Bitcoin price prediction using BERT, DistilBERT, LSTM, and XGBoost models.

## About

CryptoSynth explores a multimodal approach to forecast Bitcoin prices by leveraging:

- Sentiment analysis of approximately 150,000 Bitcoin-related tweets from the X platform.

- Transformer-based models (BERT and DistilBERT) for generating embeddings.

- Integration with historical Bitcoin prices and technical indicators.

- Advanced machine learning models including LSTM and XGBoost for prediction.

- Explainable AI (XAI) using SHAP for model interpretability.

The pipeline spans data collection, preprocessing, exploratory data analysis (EDA), sentiment classification, price prediction, and result visualization.

### Data Flow

- **Input**: Bitcoin Tweets Dataset and Tabular Data (open, close, etc.).

- **Data Preprocessing & Feature Selection**: Cleans and normalizes tweets, labels sentiments, and processes tabular data.

- **BERT/DistilBERT**: Generates embeddings from preprocessed tweets.

- **PCA**: Reduces dimensionality of combined embeddings and tabular data.

- **LSTM & Hybrid Fusion**: Applies LSTM with Random Forest and XGBoost for prediction.

- **Output**: Metrics and predictions.

## Key Features

- End-to-end sentiment analysis and price prediction.

- Integration of textual and numerical data for enhanced forecasting.

- Use of PCA for dimensionality reduction.

- Robust model evaluation with XGBoost as the best performer.

- XAI integration with SHAP for feature importance analysis.

- Comprehensive visualizations for insights.

## 🖼️ XGBoost Evaluation Metrics and XAI

### XGBoost Performance

![XGBoost Metrics](images/xgb-metrics.png)

### SHAP Value Impact

![SHAP Value Impact](images/shap-1.png)

### SHAP Model Final Output

![SHAP Final Output](images/shap-2.png)

## 📊 Performance Metrics

### On BERT Embeddings

| Model                | MSE      | RMSE    | MAE    |

|----------------------|----------|---------|--------|

| Random Forest        | 0.0001837| 0.01355 | -      |

| Random Forest (Imp)  | 0.0001095| 0.01046 | -      |

| LSTM                 | 0.007127 | -       | 0.0664 |

| Tuned LSTM           | 0.00549  | -       | 0.0576 |

| LSTM (Hybrid Fusion) | 0.0049   | -       | 0.0512 |

| XGBoost              | 0.004678 | -       | -      |

### On DistilBERT Embeddings

| Model                | MSE      | RMSE    | MAE    |

|----------------------|----------|---------|--------|

| Random Forest        | 0.0002086| 0.01444 | -      |

| Random Forest (Imp)  | 0.0001095| 0.0146  | -      |

| LSTM                 | 0.01304  | -       | 0.0937 |

| Tuned LSTM           | 0.00364  | -       | 0.0443 |

| LSTM (Hybrid Fusion) | 0.0053   | -       | 0.0531 |

| XGBoost              | 0.0047   | -       | -      |

## Setup & Usage

### Requirements

Install necessary dependencies manually:

- `torch`, `transformers`, `numpy`, `pandas`, `sklearn`

- `matplotlib`, `tqdm`, `xgboost`, `shap`

### Documentation

Refer to the following:

- [CryptoSynth_report.pdf](documentation/CryptoSynth_report.pdf): Project background, methodology, and results.

## Contributions

We welcome feedback and contributions! Feel free to fork, star 🌟, and share.

## License

Licensed under the MIT License.