Cryptocurrency Price Forecasting using Hybrid Models

Authors: Raghav Raj Sah, Kaito Hikino, Pranjal Chalise

---

Project Overview

We've built a comprehensive cryptocurrency price forecasting system that combines multiple machine learning approaches to predict Bitcoin's next-day closing price. Our system includes a baseline model, LSTM neural networks, ARIMA, XGBoost, and sentiment analysis components. The project features a complete pipeline from data preparation to model evaluation, with a focus on improving prediction accuracy through model ensembling.

---

Project Structure

.
├── Code/                             # Core implementation notebooks
│   ├── baseline.ipynb                # Simple baseline model
│   ├── lstm_model.ipynb              # LSTM implementation
│   └── milestone_1_crypto-forecast.ipynb # Data preprocessing
│
├── Data/                             # Dataset storage (not tracked)
│
├── Presentations/                    # Project presentations
│
├── submission.ipynb                  # Final submission notebook
├── paper.pdf                        # Project paper
├── SentimentAnalysis.ipynb          # Sentiment analysis implementation
├── ARIMA,_XGBoost,_Updated_LTSM,_Baseline_and_HybridModel (1).ipynb # Advanced models
├── Abstract.pdf                     # Project abstract
├── README.md                        # This file
├── requirements.txt                 # Project dependencies
└── .gitignore                       # Git ignore rules

Note: The Data/ directory is excluded from version control to keep the repository size manageable.

---

Getting Started

1. Clone and setup:

   git clone https://github.com/Khik2219/Cryptocurrency-Price-Forecasting-Using-Hybrid-Models.git
   pip install -r requirements.txt

2. Data preparation:

   • Run Code/milestone_1_crypto-forecast.ipynb to process the data

3. Model training:

   • Start with Code/baseline.ipynb for the simple model
   • Try Code/lstm_model.ipynb for the LSTM implementation
   • Explore ARIMA,_XGBoost,_Updated_LTSM,_Baseline_and_HybridModel (1).ipynb for advanced models
   • Check out SentimentAnalysis.ipynb for sentiment-based predictions

4. Final results:

   • See submission.ipynb for the complete implementation and results
   • Read paper.pdf for detailed methodology and findings

---

What's Inside

• Data Processing (milestone_1_crypto-forecast.ipynb): Handles data cleaning, feature engineering, and train/test splitting
• Baseline Model (baseline.ipynb): Implements a simple "tomorrow equals today" prediction
• LSTM Model (lstm_model.ipynb): Deep learning approach using 30-day sequences
• Advanced Models (ARIMA,_XGBoost,_Updated_LTSM,_Baseline_and_HybridModel (1).ipynb): Combines multiple models for better predictions
• Sentiment Analysis (SentimentAnalysis.ipynb): Integrates social media sentiment data
• Final Implementation (submission.ipynb): Complete solution with all components

---

Key Features

• Multiple model approaches (LSTM, ARIMA, XGBoost)
• Sentiment analysis integration
• Model ensembling
• Comprehensive evaluation metrics
• Clean, modular code structure

---

Notes

• The Data/ directory is excluded from git to keep the repository size manageable
• External datasets should be cited according to their original sources
• All code is well-documented with clear explanations

---

License

This project is for academic purposes. External code and data are subject to their respective licenses.

---

Contact

Feel free to reach out through GitHub for questions or collaboration.

---

GitHub Repository: [https://github.com/Khik2219/Cryptocurrency-Price-Forecasting-Using-Hybrid-Models.git]

---