README.md

Project Goldendwarf

Utilizing machine learning and natural language processing techniques to predict future price movements of cryptocurrencies (and stocks) as well as gauging social sentiment of the crypto universe.

Purpose

• Applying neural network fundamentals & data analytics knowledge to forecast price action in the cryptocurrency space.
• Outputs from analysis would collectively help determine potential market risk in the buying/selling of cryptocurrency assets.
• Looks to the future would be directed toward using this collective analysis to feed a real-time automated machine that generates alpha through trading cryptocurrency.

Questions to Answer

• Can the model accurately predict price movement more than 50% of the time? (Is it better than flipping a coin?)
• Can the contributive analysis be successful in feeding information to a well-operating trading bot (for future development)?
  • Can the code be optimized sufficiently to be real-time operational (or close to it)?
  • Is our implementation strategy scaleable?

Analysis

Components

• Machine Learning (See the neural_network folder for further description & code):

  • FB Prophet: Using Facebook's FBProphet machine learning library to forecast crypto price-movement in real-time.
  • Bi-Directional RNN: Training our own bi-directional recurrent neural network for greater improvements in predictive accuracy (currently in production).

• Natural Language Processing (See the NLP_analysis folder and NLP Wordcloud folder for further description & code):

  • Scraping and analyzing relevant twitter data to develop a cryptocurrency social sentiment score in real-time.

Data & Data Storage

• Data scraped using the following APIs:
  • Binanace API
  • Tweepy Library (Twitter API)
• Database
  • MongoDB for data storage (for future referencing and retrieving relevant tables for the interactive dashboard).
  • Data is updated every time a new coin is selected.

See sample price-action data stored in MongoDB below (visualized as dataframes using Streamlit)

Dashboard

• Check out our Demo Video to see how the interactive dashboard operates.
• Created using Streamlit and Plotly, the interactive dashboard displays price forecasting of selected coins and social sentiment scores of the cryptocurrency market in real-time.
• Code can be referenced in the Dashboard folder.

Google Slides

• Follow this link to see our presentation slides.