Forecast a time series using multiple models — forecast

Tests multiple forecasting models, selects the best one based on accuracy metrics, and also provides an ensemble (average) forecast.

Usage

forecast_series(
  data,
  time_col = "tempo",
  value_col = "valore",
  freq_col = "FREQ",
  horizon = NULL,
  models = c("auto.arima", "ets", "theta", "naive", "snaive"),
  metric = c("RMSE", "MAE", "MAPE"),
  test_size = 0.2,
  save_path = NULL,
  verbose = TRUE
)

Arguments

data: data.table with a single time series
time_col: Column name for dates (default "tempo")
value_col: Column name for values (default "valore")
freq_col: Column name for frequency indicator (default "FREQ")
horizon: Forecast periods. NULL for auto-detect (24 monthly, 8 quarterly, 2 annual)
models: Character vector of model names to test
metric: Accuracy metric for model selection: "RMSE", "MAE", or "MAPE"
test_size: Proportion of data to use for testing (default 0.2)
save_path: Optional file path to save the forecast as RDS
verbose: Print progress messages (default TRUE)

Value

An S3 object of class "istat_forecast" containing:

best_model: Best performing model with forecast and accuracy
ensemble: Average forecast from all models
all_models: Results from each individual model
metadata: Forecast parameters and metadata
original_data: Input data

Examples

if (FALSE) { # \dontrun{
# Load some ISTAT data
data <- download_istat_data("534_50", start_time = "2015")
data <- apply_labels(data)

# Filter to a single series
series <- data[ECON_ACTIVITY_NACE_2007_label == "TOTALE INDUSTRIA E SERVIZI  (b-n)"]

# Generate forecast
fc <- forecast_series(series, horizon = 24)

# View results
print(fc)
plot(fc)
} # }