End-to-end ML data product predicting next-day high prices for four cryptocurrencies, with an LSTM model for Ethereum achieving a test MAE of 70.8, served via FastAPI and visualised in a Streamlit dashboard with live market data.
Built as a group project for the Advanced Machine Learning Applications subject at UTS, this data product predicts next-day high prices for Bitcoin, Ethereum, XRP, and Solana. Each team member was responsible for one cryptocurrency. For Ethereum, five experiments were run progressively, starting from baseline models (Linear Regression, Random Forest) through to an LSTM neural network that significantly outperformed classical approaches, confirmed by statistical tests (p < 0.001). The final LSTM model uses a sequence of 5 timesteps, 64 units, and was trained with the Adam optimizer using MAE as the loss function. It achieved a training MAE of 23.4, validation MAE of 90.8, and test MAE of 70.8, demonstrating consistent generalisation across all splits. Live market data and news summaries are pulled from the CoinGecko API and displayed alongside predictions in a Streamlit dashboard. The full pipeline, including the trained model and preprocessing artifacts, is served via a FastAPI endpoint deployed on Render.
