Cryptocurrency Trend Predictor
A time-series forecasting model for cryptocurrency price trends, built with Facebook Prophet.
The Problem
Crypto price series are highly volatile with irregular seasonality, which breaks a lot of classical forecasting assumptions.
Why I Built It
Wanted to evaluate how well an additive time-series model built for business forecasting (holidays, weekly/yearly seasonality) holds up on a series with none of Prophet's usual assumptions about regularity.
What I Built
Pulled historical price data, configured and tuned a Prophet model (trend changepoints, seasonality components), and evaluated forecast accuracy against a holdout period.
Technical Challenges
Prophet's seasonality assumptions don't map cleanly onto crypto markets, so a meaningful part of the work was tuning changepoint sensitivity to avoid the model either overreacting to noise or missing real trend shifts.