mmsaki / price_prediction_models

We use machine learning tools to predict the price of ethereum from historical data, economic indicators, and community sentiment on ethereum specifically from twitter.

Group Project 2

Table of Contents

1. Hypothesis
2. Data Collection and Clenaing
3. Choices for models & Perfromance
4. Contributions

Hypothesis

• We use machine learning tools to predict the price of ethereum from historical data, economic indicators, and community sentiment on ethereum specifically from twitter.
• We test this hypothesis by building LSTM and GRU models

Data Collection & Cleaning

• Our Sources were:
  • Twint Protocol for Collecting tweet data
  • OWlracle API for collecting Gas price history
  • Kaggle for ETH to USD Historical Data
  • FRED for personal savings percentage data
  • Market Watch for S&P 500 historical data

Choices for models & Performance scores

• We used two models to predict the price of ethereum

  • Long Short-Term Memory (LSTM) model from keras
  • Gated Recurrent Unit (GRU) model form keras

• Spliting our data to train and test

• LSTM vs GRU model perfromances

  • LSTM evaluation had 8.98% loss, 91% accuracy

  • GRU evaluation had 10.65% loss, 89.35% accuracy

Contributions

• Meek Msaki

  • Get historical gas prices and clean data.
  • Set up and run LSTM model

• Kyle Plathe

  • Get historical eth prices and clean data
  • Get S&P 500 historical
  • Get US savings historical data
  • Set up GRU model

• Richard Melvin

  • Get historical sentiment for ethereum 2017 - 2020
  • Twitter api sentiment analysis with nltk and vader
  • Perfromed PCA analysis

Presentation & Files

Files:

• Gas price data collection

• Twint Protocal data collection

• Eth data cleanup

• Sentiment Analysis

• LSTM Model price predictions

• GRU Model price predictions

• Project Presentation