BTC-Price-Prediction-ML-Project

About

This project focuses on the prediction of the prices of Bitcoin, the most in-demand crypto-currency of today’s world. We predict the prices accurately by gathering data available at coinmarketcap while taking various hyper-parameters into consideration which have affected the bitcoin prices until now.

Paper presentation

The paper contains all details of algorithms used along with results, anaylisis and discussions om the topic.

Dataset

• Dataset has been downloaded using coinmarketcap API.

• Dataset after Preprocessing

• Seasonal Decomposition of the Time-Series after order-1 differencing (to make it stationary)

• Correlation Plots

• Best Results (GARCH + SARIMAX) (RMSE: 154.32)

ML Models Used:

• Regression Models
  • Linear Regression with various penalties
  • Polynomial Regression
  • Bayesian Regression
• ARIMA Models
  • AR
  • ARMA
  • ARIMA
  • SARIMAX
• SARIMAX + GARCH on the residuals of SARIMAX model
• VAR Model

Python Dependencies:

• pandas
• numpy
• requests
• matplotlib
• statsmodels
• pmdarima
• arch

Install Dependencies (requirements.txt)

1. pip install -r requirements.txt

OR

1. pipenv install --ignore
2. pipenv shell

How to Run

1. cd <PROJECT ROOT DIRECTORY>
2. python <filename>.py

File Descriptions:

• auto-ARIMA.py: Runs automated gridsearch from pmdarima library, to find the best model parameters.
• AR.py, ARMA.py, ARIMA.py, SARIMAX.py use the above found best parameters to train the respective models as per their filenames.
• GARCH-SARIMAX.py runs SARIMAX models added with error of residuals from SARIMAX using GARCH.
• elasticnet.py runs Linear Regression with a combination of L1 and L2 penalty.
• bayesian.py runs BayesianRidge regression with optimal parameters.
• polyreg.py runs Linear Regression by adding polynomial features.
• var.py runs runs VAR model on the data.