CoinSeer

• 

  Prediction of XRP

• 

  Prediction of Bitcoin based on a model trained on data till today

• 

  Prediction of Bitcoin based on a model trained on data till 9 days back

• 

  Architecture part 1

• 

  Architecture part 2

Inspiration

Cryptocurrencies have taken over the world..yet again and this time it looks serious. With their prices skyrocketing, it is important for investors to stay one step ahead of everyone else. So in this project I built a tool that constantly keeps track of the trends using machine learning.

What it does

CoinSeer predicts the crypto prices. It takes into account everyday price changes and learns from it to predict future prices.

How we built it

• I built a crawler that reads historical data of the market trends from coinmarketcap.com and it is capable of reading new data everyday. For demo purposes I have picked 2 cryptocurrencies bitcoin and XRP.
• This data is then preprocessed. The dataset is windowed over time and the next value after the window is what we intend to predict. I do this over the entire dataset to create training data.
• I trained an LSTM based nueral network using this dataset.
• I then created a flask based rest service that can serve as an interface between the ML model and the Appsmith UI.
• Appsmith helps us in visualising the results by providing easy UI tools that can talk to webservice seamlessly. I display graphs taht show trends of both history and predictions and also a tabular data for closer analysis. UI also helps users to swicth between different crypto currencies and hence different models in the backend.
• To make it more realistic, I trained multiple models using datasets till today(1), till yesterday(2), day before yesterday(3), .. 9 days before(10). This way the prediction of the crypto prices are more robust.

Challenges we ran into

• Coinmarketcap actively hinders crawling. I had to get innovative.
• Crypto currency prices are very volatile making training models hard.
• Appsmith had a bit of a learning curve, but I got used to it pretty quickly
• Handling graphs was a bit challenging

Accomplishments that we're proud of

This project involves web crawling, machine learning, REST backend and front end. It feels good to have touched so many areas.

What we learned

Learnt how to build UI with Appsmith quickly. Learnt about Flask and web crawling

What's next for CoinSeer

• Train for other crypto currencies.
• Host CoinSeer on cloud.
• Find better nueral architecture that considers more than historical data.

Built With

• appsmith
• python
• tensorflow