Glosarium AI
Kamus lengkap Kecerdasan Buatan
SARIMA
Extension of the ARIMA model incorporating seasonal components (S) to capture periodic patterns in time series with regular seasonal variations.
Stationarity
Statistical property where the mean, variance and autocorrelation structure of a time series remain constant over time, an essential condition for applying ARIMA models.
Differencing
Mathematical transformation consisting of subtracting each observation from the previous one to eliminate trends and convert a non-stationary time series into a stationary one.
Autocorrelation (ACF)
Measure of correlation between an observation and previous observations at different time lags, used to identify MA orders in ARIMA models.
Partial Autocorrelation (PACF)
Correlation between an observation and its time lags after removing the effects of intermediate lags, essential for determining the AR order in ARIMA models.
Order of Integration (d)
Parameter indicating the number of differencings required to make a time series stationary, representing the 'I' component in the ARIMA(p,d,q) model.
Autoregressive Order (p)
Parameter specifying the number of autoregressive terms in the ARIMA model, representing the dependence of the current value on previous values.
Moving Average Order (q)
Parameter defining the number of moving average terms in the ARIMA model, representing the dependence of the current error on previous forecast errors.
Seasonality
Repetitive and predictable pattern in a time series over fixed time intervals (monthly, quarterly, yearly), modeled by the SARIMA(P,D,Q)s components.
Bayesian Information Criterion (BIC)
Model selection criterion similar to AIC but with a stricter penalty for complexity, favoring more parsimonious models in ARIMA selection.
White Noise
A sequence of independent and identically distributed random variables with zero mean and constant variance, representing the residual error in valid ARIMA models.
Augmented Dickey-Fuller Test
A statistical test that checks for the presence of a unit root in a time series to determine its stationarity and justify the need for differencing in ARIMA models.
Seasonal Autocorrelation Function
Seasonal version of the ACF that measures the correlation between observations separated by multiples of the seasonal period, used to identify SARIMA orders.
Out-of-Sample Forecasting
Evaluation of the model's predictive performance on data not used during training, measuring the generalization ability of ARIMA/SARIMA models.
Backtesting
Historical validation procedure that tests the predictive performance of ARIMA models by simulating predictions on past periods to evaluate their robustness.