Bitcoin Price Forecasting Using the Facebook Prophet Model for Accuracy Analysis and Crypto Market Trends

This study aims to evaluate the performance of the Facebook Prophet model in forecasting Bitcoin prices, a cryptocurrency known for its high volatility and non-stationary characteristics. The model is configured using multiplicative seasonality, a changepoint_prior_scale value of 0.1 to maintain sensitivity to trend changes, and an added monthly seasonal component with a period of approximately 30.5 days to capture short-term cycles not detected by weekly or yearly seasonality. Historical data is preprocessed through light smoothing to reduce extreme noise without eliminating trend signals, followed by training and forecasting over a 180-day horizon. Performance evaluation using MAE, RMSE, and MAPE metrics yielded values of 2,647.99 USD, 4,178.26 USD, and 17.73% respectively, indicating adequate accuracy for a highly volatile asset. Decomposition analysis reveals a strengthening long-term trend since 2024, a weekly pattern that peaks midweek, and a yearly pattern that weakens in early September and strengthens toward the end of the year. Forecast results show an upward trend tendency with increasing uncertainty over time, confirming Prophet’s role as an interpretable and efficient baseline model for short- to medium-term Bitcoin price forecasting.