SC1015 Data Science Project — Crypto-Genie

This project aims to effectively mitigate the volatility of cryptocurrency investment by evaluating the media hype from Google Trends in our prediction models.

---

🛠️ Installation and Set Up

• Clone repository

  git clone https://github.com/crystalcheong/crypto-genie.git
  

• Install dependencies with pip

  $ pip install -r requirements.txt
  

---

📂 Project Structure

📦crypto-genie
 ┣ 📂data
 ┃ ┣ 📂searchTrends
 ┃ ┣ 📜BTC-SearchTrend.csv
 ┃ ┗ 📜README.md
 ┣ 📂metrics
 ┃ ┣ 📜README.md
 ┣ 📂models
 ┣ 📜0_DataScraper.ipynb
 ┣ 📜1_DataAnalysis.ipynb
 ┣ 📜2_UnivariateForecast.ipynb
 ┣ 📜3_MultivariateForecast.ipynb
 ┣ 📜README.md
 ┗ 📜requirements.txt

/data - stores all the collected data to be utilized
/metrics - contains the exported measurement of accuracy & efficacy
/models - contains the exported pre-trained models

---

📑 Data Sources

• Financial Data - Yahoo Finance / yfinance
• Search Trends Data - Google Trends / pytrends

---

🧭 In-depth Documentation

• Data Curation & Preparation
• Evaluation of Model Performance

---

📚 Notebooks Overview

Each notebook is prefixed with the chronological order of the analysis pipeline and can be executed as a standalone.

• Data Scraper

  • Retrieves stock information on Bitcoin (BTC-USD) from Yahoo Finance
  • Annual archive of Google Search Trends using the pytrends package
  • Generate and export BTC-SearchTrend.csv from the successful merger of above-mentioned datasets

• Data Analysis

  • Displays dataset overview and checks for any null values
  • Exploratory data analysis on the BTC-SearchTrend.csv dataset

• Univariate Forecast

  • Utilises one stock ticker variable (OPEN / CLOSE / HIGH / LOW) as the model data
  • Initialise, train and predict Bitcoin (BTC-USD) valuation with the following machine learning model(s):
    • Rolling-Forecast ARIMA
    • XGBRegressor
    • LSTM
  • Summarises and compares the performance of all previously ran models

• Multivariate Forecast

  • Utilises all stock ticker and search trend variables as the model data
  • Initialise, train and predict Bitcoin (BTC-USD) valuation with the following machine learning model(s):
    • LSTM
  • Summarises and compares the performance of all previously ran models

---

🗝️ Project Takeaways

In conclusion, the inclusion of media hype from Google Trends resulted in a more accurate multivariate forecasting. That said, as the Google Trends search percentile is not the only definitive measurement of media hype, the accuracy and efficacy of the prediction models can be further enhanced by the integration of a multi-faceted data curation from other sources such as the users' sentiments from Reddit and Twitter

• Learning Outcomes
  • Interpret stock information
  • Develop & evaluate time series forecasting machine learning models such as
    • Rolling-Forecast ARIMA
    • XGBRegressor
    • LSTM
  • Utilized external Python libraries such as plotly, statsmodels and tensorflow
  • Collaboration on Google Colab and git repository management with Github

---

🧰 Languages & Tools

• Languages
  

• Libraries, Packages
  

• Tools, IDE
  

---

Contributors ✨

Crystal Cheong	Yue Chong	Jared Chan
Data Collection & Preparation Data Analysis Machine Learning Repository Documentation Presentation Slides & Script	Data Analysis Data Visualisations Presentation Slides & Script Video Presenter	Machine Learning Presentation Slides & Script Video Presenter

---

💡 References

• ARIMA Model for Time Series Forecasting
• Forecasting BTC
• XGBoost for Time Series Forecasting Intro to Long Short-Term Memory (LSTM)
• LSTM Models for Time Series Forecasting
• Multi-Step LSTM Time Series Forecasting Models
• Import & Concat multiple CSV

This repository is submitted as a project work for Nanyang Technological University's SC1015- Data Science and Artificial Intelligence course.