B”H
← Back to Home
Bitcoin Price Movement Prediction Using Google Trends
Research Question
Can Google Trends search data for cryptocurrency-related keywords be used to predict next-day Bitcoin price movement (up or down) with better-than-random accuracy?
Project Overview
This project explores the relationship between public interest in cryptocurrency (measured through Google Trends) and Bitcoin price movements. The goal is to develop a binary classification model that predicts whether Bitcoin’s price will increase or decrease the following day, providing actionable insights for short-term trading strategies.
The analysis encompasses three major components:
- Data Collection & Normalization - Aggregating 10 years of daily Google Trends data for 5 cryptocurrency-related search terms (“bitcoin wallet”, “crypto wallet”, “buy crypto”, “bitget”, “buy bitcoin”) using proxy rotation to bypass API limitations, and normalizing data across overlapping time windows
- Feature Engineering & EDA - Cleaning and merging Bitcoin historical price data with Google Trends data, engineering technical indicators (RSI, MACD, Bollinger Bands), momentum features, rolling averages, and creating a binary target variable for next-day price direction
- Model Building & Evaluation - Developing XGBoost classification models with GPU-accelerated hyperparameter tuning, comparing performance against naive baseline models, and evaluating predictive accuracy on held-out test data
The project demonstrates advanced data engineering techniques including proxy configuration for large-scale web data collection, time series normalization across discontinuous data windows, multivariate feature engineering, and machine learning model optimization for financial prediction tasks.
Project Files