Automatic ARIMA model selection using auto.arima with STL decomposition, multiple stationarity tests, and adaptive forecast limits.
Uses auto.arima with adaptive constraints based on data size and frequency.
Seasonal-Trend decomposition using Loess for trend, seasonal, and remainder components.
ADF, KPSS, Phillips-Perron tests plus Ljung-Box for residual autocorrelation.
Provide a dataset with time_column and value_column. Specify frequency (hourly, daily, weekly, monthly, quarterly, yearly).
Automatically limits forecast periods based on data length and frequency to ensure reliability. Requires minimum observations based on frequency (e.g., 24 months for monthly data). Handles multiple date formats with intelligent parsing.
From stationarity to validated predictions
ADF, KPSS, and Phillips-Perron tests to assess need for differencing.
auto.arima with adaptive max orders based on data size and seasonal detection.
STL decomposition, generate forecasts with confidence bands, Ljung-Box test.
Intelligent ARIMA forecasting with automatic safeguards—prevents overfitting by limiting forecast horizons based on data characteristics.
The implementation uses auto.arima from the forecast package, automatically adjusts model complexity for large datasets (stepwise for n>500), limits forecast periods to maximum 50% of historical data, performs STL decomposition for seasonal patterns, and includes comprehensive diagnostics with ACF/PACF plots, residual analysis, and multiple stationarity tests.
Note: Forecast periods automatically capped based on frequency (e.g., max 365 days for daily, 60 months for monthly). Uses stepwise selection for efficiency with large datasets.
Generate calibrated predictions with diagnostics
Read the article: ARIMA Forecasting