Time Series: ARIMA/SARIMA (Box-Jenkins)¶
Complete time series analysis following the Box-Jenkins methodology, from model identification to forecasting with confidence intervals.
Contents¶
The dashboard presents 6 interactive tabs:
| Tab | Content |
|---|---|
| Original Series | Monthly airline passengers 1949-1960 (144 observations) |
| Decomposition | Trend + Seasonality + Residual (multiplicative) |
| ACF / PACF | Autocorrelation and partial autocorrelation of the differenced series |
| Diagnostics | Residuals, histogram, residual ACF, Q-Q plot |
| Final Forecast | Original series + SARIMA fit + 12-month forecast + 95% CI |
| Radar Metrics | RMSE, MAE, MAPE, R2 of the model |
Box-Jenkins Methodology¶
- Identification: ACF/PACF to determine orders (p, d, q)(P, D, Q)[s]
- Estimation: Maximum likelihood fitting
- Diagnostics: Ljung-Box and Jarque-Bera tests on residuals
- Forecasting: Forecast with 95% confidence intervals
Selected model: SARIMA(1,1,0)(0,1,0)[12] -- AIC = -445.41
---
Source Code¶
- Full script:
ejercicios/04_machine_learning/07_series_temporales_arima/ - Theoretical guide:
07_series_temporales_arima/README.md