Predicting Cryptocurrency Prices

About

This is our Mini-Project for SC1015 (Introduction to Data Science and Artificial Intelligence) which aims to predict cryptocurrency prices of popular cryptocurrencies, scraping data from Yahoo-Finance

Contributors

• @NotAnAddictz - Data Extraction, Tensorflow, FbProphet
• @weix1233 - FbProphet, Tensorflow
• @avdheshcharjan - Data Visualisation, Data Extraction

Main Packages Used

• pandas_datareader
• fbprophet
• matplotlib
• tensorflow
• seaborn
• sklearn

Datasets Used (Scraped from Yahoo Finances)

• AAVE (AAVE)
• BinanceCoin (BNB)
• Bitcoin (BTC)
• ChainLink (LINK)
• Cardano (ADA)
• Ethereum (ETH)
• Solana (SOL)
• Tether (USDT)
• Uniswap (UNI1)

Defining the Problem

• Are we able to accurately predict the future prices of the various cryptocurrencies?
• Which machine-learning model is more accurate in their prediction?

Models Used

• Tensorflow LSTM
• FbProphet

Conclusion

• Changes in cryptocurrency prices are not reliant on other cryptocurrencies
• Cryptocurrencies are generally predictable as long as there is no sudden changes that affects their prices significantly
• Future prediction will have a wider range the further one predicts
• Having a larger (and more recent) dataset will result in higher prediction accuracy
• FbProphet seems to be clearer and more accurate in predicting various cryptocurrencies

What did we learn

• Scraping datasets using pandas datareader
• FbProphet and its implementation
• Neural Networks, Keras and Tensorflow
• Data Visualisation

References

• https://github.com/rohan-paul/Cryptocurrency-Kaggle
• https://towardsdatascience.com/cryptocurrency-price-prediction-using-lstms-tensorflow-for-hackers-part-iii-264fcdbccd3f
• https://facebook.github.io/prophet/docs/quick_start.html