Bitcoin Price Prediction Using Twitter Sentiment Analysis

Team Members

Abhineet Gupta

Jigar Soni

Nikita Sengupta

Vineet Zunjarwad

Setup for collecting older tweets

To bypass the limitation of the twitter API, that is to get tweets more than 2 weeks earlier, we have scrapped the original twitter web page using a JSON provider. By which we can use infinite scrolling and get the previous data directly.

1. $ pip install -r requirements.txt

2. $ python Exporter.py --querysearch "europe refugees" --maxtweets 10000

Setup for running prediction script

1. Create a new virtual environment

   $ virtualenv -p python3 venv

   $ source venv/bin/activate

   $ pip install -r requirements.txt

2. To calculate algorithm accuracy you will need feature.csv that you cann be found here

   $ python predict.py

Algorithm accuracy

We have acheived Mean Square Error close to 33 over the span of roughly 650K records with 70-30 split using Logistic Regression model.

Final Output:

The final out of the analysis is a website which shows signals: (To buy or not to buy botcoins on particular day!) This front-end server is a node JS server, deployed on Heroku. The website link is: https://vast-headland-21279.herokuapp.com/

The signals are being calculated using peak/valley values and slope of the bitcoin stock price at that interval.

The outcome of our project:

Here 1 suggests buying signals, -1 suggests selling signals and 0 suggests to retain whatever amount you have!