Kaggle Crypto Forecasting
Machine-learning price prediction for 14 cryptocurrencies.
Jan 2022
Built for the G-Research Kaggle crypto forecasting competition, this project predicts short-term returns for 14 cryptocurrencies — Bitcoin, Ethereum, Binance Coin, Dogecoin, and more — using gradient-boosted decision trees.
I focused on data collection and feature engineering across a wide range of sources, from S&P 500 market data to on-chain whale transactions, compiling everything into a clean master dataset used to train the models. The final XGBoost models scored roughly 0.47 correlation for Bitcoin, with the other coins landing in the 0.2–0.4 range.
Key features
14-coin return prediction
Per-coin XGBoost models forecasting short-term returns across the major cryptocurrencies.
Multi-source feature engineering
Engineered features spanning traditional markets (S&P 500) to on-chain whale-transaction data.
Unified training pipeline
Collected and compiled disparate sources into a single clean master dataset for model training.
Collaborative + teaching
Worked on a team, helping teach feature engineering in R and Python to improve the group's estimates.