End-to-end ML data product predicting next-day high prices for Bitcoin, Ethereum, XRP, and Solana. Individually responsible for the Ethereum LSTM model and the CoinGecko data pipeline. Served via FastAPI and visualised in Streamlit.
A group ML data product built as part of the Advanced Machine Learning Applications subject at UTS, predicting next-day high prices for four cryptocurrencies: Bitcoin, Ethereum, XRP, and Solana. I was individually responsible for the Ethereum prediction model and the shared data infrastructure. For Ethereum, I ran four experiments evaluating Linear Regression, Random Forest, ElasticNet, HistGradientBoosting, and LSTM architectures. LSTM significantly outperformed classical models (validation MAE of 118.3 vs 1,301.9 for HGB), and the final production pipeline uses an LSTM with timesteps=5, units=64, lr=0.001, batch size=32, achieving a test MAE of 70.78. I also built the data pipeline and news feed powered by the CoinGecko API, which supplies live price data and market summaries to the dashboard. All models are served via FastAPI with the .keras model and preprocessing artifacts (MinMaxScaler, SequenceTransformer) exported to the /models directory. The Streamlit dashboard visualises predictions alongside live market context. Deployed on Render.
