ARIMA forecasts time-series values from prior observations by combining autoregression, differencing, and moving-average error terms. It is often an early forecasting choice alongside Exponential Smoothing, while its forecast errors can reveal unusual observations. section 17.2.6
A practical workflow converts timestamped tabular data into a Darts TimeSeries, fits ARIMA on an initial segment, predicts a held-out horizon, and compares predictions with actual values. Parameters require experimental tuning; forecasts may preserve weekly patterns but increasingly rely on the mean farther into the future. section 17.2.6
Forecasting a held-out horizon
Synthetic daily data are generated from the previous value, weekday, and randomness. With ARIMA(p=7, d=2, q=2), the first observations are used for fitting and the final 60 values are predicted. A deliberately enlarged value produces a substantially larger-than-usual forecast error, supporting outlier detection. section 17.2.6
Listing 17.7 Forecasting values using Darts with ARIMA
from darts import TimeSeries from darts.models import ARIMA df['timestamp'] = df.index #1 series = TimeSeries.from_dataframe(df, "timestamp", "Values") model = ARIMA(p=7, d=2, q=2) #2 model.fit(series[:-60]) pred = model.predict(60) #3 pred.plot(label="forecast", low_quantile=0.05, high_quantile=0.95) #4 plt.legend() plt.show()
Inspect the concise Darts implementation for constructing a TimeSeries, fitting ARIMA, predicting 60 points, and plotting forecast quantiles. section 17.2.6
