https://docs.google.com/presentation/d/11-A77KxuFXHH2xDc9bvrYPjjifkujIl8LBE7ybXscUk/edit?usp=sharing
For our Columbia University Data Analytics Bootcamp Project our group analyzed cryptocurrencies in two stages to help predict the future price. The first stage was an analysis of the Top 5 most traded coins from 2016-2021, this was to educate both our group and the reader of the financial environment among cryptocurrencies. The second stage was creating a Machine Learning model that would predict these coins prices for the next 30 days.
The data for our Analysis portion of the project came from Facebook Prophet. Facebook Prophet is an opensource software that also provides relatively clean data for cryptocurrencies. The software can help forecast a time series, its built-in models are mainly used to recognize yearly, weekly, and daily seasonality; we used this as an additive to our Machine Learning model. Through FB Prophet and Plotly we were able to create visualizations that helped breakdown the cryptocurrencies market and its activity, like the one below that shows the volume of shares for the top traded coins since 2016. The rest of these visualizations are included in our Google Slides presentation, linked above.
To create a machine learning model, we first began by sourcing Bitcoin pricing data from Yahoo Finance Live Data. This data was minute over minute, because the amount of data that needed to be trained, we used the most recent 3 months. Python and Sklearn was used to train/test our cleaned 3-month pricing data. Once the data was trained it was run through a Sklearn SVR (Support Vector Regression) model which was then altered to export a 30-day prediction for Bitcoin pricing
PostgresSQL was used to store and manipulate data provided by FB Prophet before being imported to Jupyter Notebooks for visualizations.
Dashboard is hosted by Dash, it shows an interactive graph of the top 12 cryptocurrencies dating from 2016-Present. Our machine learning model is implemented to display a 30-day prediction for the top 5 cryptocurrencies (Bitcoin, Ethereum, Ripple, Litecoin and Tether) based on static data. Multiple coins can be displayed at once and selected areas of the graph can be enlarged with adapting axes.
