AI 용어집
인공지능 완전 사전
Time Series Decomposition
Statistical process consisting of separating a time series into its fundamental components (trend, seasonality, residuals) to better analyze its structure.
Changepoint
Point in time where the growth rate of the trend changes, automatically detected by Prophet or manually specified to improve model accuracy.
Holiday Effects
Component of the Prophet model capturing abnormal and recurring variations related to specific non-regular calendar events.
Piecewise Linear Trend Regression
Trend modeling method in Prophet where the curve is approximated by a succession of linear segments with variable slopes at changepoints.
Fourier Series
Mathematical representation used by Prophet to model seasonal effects, approximating cyclical patterns by a sum of sinusoidal functions.
Trend Flexibility Adjustment
Parameter 'changepoint_prior_scale' controlling trend flexibility, allowing adjustment of the model's sensitivity to growth changes.
Uncertainty Intervals
Ranges of probable values around forecasts, generated by Prophet by simulating thousands of possible futures to quantify forecast risk.
MARS (Multivariate Adaptive Regression Splines)
Non-linear modeling technique inspiring Prophet's approach for additive decomposition and management of non-linearities in time series.
Stan
Probabilistic programming platform used by Prophet for Bayesian inference, enabling estimation of model parameters and their uncertainty.
Growth Saturation
Prophet feature that models phenomena where growth slows down and reaches an upper limit, using logistic functions.
Exogenous Regressors
Additional external variables integrated into the Prophet model to improve forecast accuracy by capturing external influencing factors.