• Team name: It's not a bug, it's a feature
  • Category: Economics
  • Datasets (Codejam): Cryptocurrencies
  • Datasets (External): Data scraped from social media / news


  • $$$ for bubble tea

What it does

  • Cryptocurrency forecasting
  • News and Social Media Scraping
    • Sentiment Analysis (Invest, or not to invest?)
    • Analyzing correlation between social media activity and cryptocurrency price
  • Dashboard to visualize results

How we built it

  • Frontend
    • Used Angular 5 to build a cross-platform application with Material Design
  • Backend
    • Built an API with Flask and Flask-RESTful
  • Cryptocurrency Forecasting
    • Used scikit-learn, to train a model which can predict future cryptocurrency prices,
  • News / Social Media Scraping
  • Scripts developed in Python
  • Sentiment Analysis
    • Used TextBlob to determine the polarity and subjectivity for a given news article, or block of text. From this, we categorized the text as either positive, neutral, or negatively correlated with cryptocurrency price and trends
  • Social media analysis
    • Used Facebook's developer api tools along with python to scape data from groups, along with other information about the posts made on each post for further analysis.

Challenges we ran into

We experimented with many different approaches to forecasting future prices, and eventually concluded with predicting a future rolling mean value. We found price forecasting in itself is almost impossible, only being able to forecast the price at an accuracy not much better than random. In the end, we found that predicting some sort of averaged value or "trend" was much more feasible.

Overall, having enough time to build a fully functional frontend which operates in real time was a challenge. For the frontend itself, it was a challenge to integrate Material Design with Angular 5 since the updates aren't always compatible. Second, data visualization is time-consuming but we managed to graph the most crucial information on the dashboard.

Given all the collected information in terms of the sentimental analysis and social media analysis, integration was another challenge. If we were able to train the forecasting with the other training features, the prediction could be improved drastically.

Accomplishments that we're proud of

We're alive. Frontend works with Material Design. Sentiment Analysis work great. Cryptocurrency price analysis works amazing. Trend tracking with social media works too. Things are semi-connected. The API calls for the frontend exists even though it's not connected YET. But really though, we're alive.

What we learned

Much machine learning. Such data. Wow cryptocurrency.

What's next for Cryptocurrency Dashboard

Integrate all the data together to predict future prices with price history and social media analysis together.

Share this project: