# Automated model selection in forecasts

Multiple model types are used to create candidate models for each time series in a forecast. All available model types are normally used, except when a seasonal component is absent. There are only three model types available that do not account for seasonality in the data.

The default value of Auto for the Seasonal period option detects seasonal period length through spectral analysis. A model is then trained using the period detected through spectral analysis.

Multiple models are compared by using a model error and the number of model parameters. For example, when model errors are equal for two models, the model with fewer parameters is preferred. The latter model provides a more condensed representation of the observed data and also tends to generate more reliable forecasts.