### Forecast

Pick a forecasting model and train it to predict future values of a numeric column.

Pick a forecasting model and train it to predict future values of a numeric column.

### Note

The speed and quality slider is more of a spectrum. This setting determines the number of machine learning models considered. Models that do not support the training data are automatically excluded.

When you run the Forecast tool, you make a model capable of in-sample predictions. Varicent ELT then creates predictions for the dates in the available data. Finally, using both the actual and the predicted data, Varicent ELT computes the error score. For more information about error scores, read: Scores.

In the Forecast tool configuration pane, enter the following information for each field:

Field

Description

Forecast type

Select the forecast type to use. If you are unsure which type to use, select

**Automatic**, which chooses the best available forecasting model. Learn more...Forecast value column

Select the column from your dataset to use to forecast.

Date column

Select the date column to use.

Frequency

Select the frequency of the time period, such as daily, monthly, yearly and so on.

(Advanced) Use multiple columns

Turn on to use the multivariate forecast. Learn more...

(Advanced) Static attributes

Select the columns to use with static attributes. This is useful when using the For Each feature.

(Advanced) Exclude columns

Select the columns to exclude from the forecast.

(Advanced) Smart Exclude

Select the checkbox to have the tool exclude columns from the forecast.

(Advanced) New column name

Enter a name for the new column.

(Advanced) Number of periods

Select the number of periods to use.

(Advanced) Speed/Quality slider

Use the slider to select using speed or quality for the output of your forecast.

(Advanced) For Each

Optionally select for more control over how Varicent ELT forecasts your data. Learn more...

(Advanced) Ignore subgroup limit

Select to override the limit of 10 groups, and allows any number of forecasts to be built. Learn more...

The tool is configured.

#### Forecasting models

Automatic: Picks the best available forecasting model. It trains the model to predict future values of a numeric column.

ARIMA: Short for "Auto Regressive Integrated Moving Average". It applies Auto Regressive and Moving Average models to time series data to predict the future.

Linear Regressive: Identifies underlying trends. It draws a straight line through data points to minimize the distances between data points and the resulting trend line.

Prophet: Forecasts time series data by incorporating non-linear trends, weekly, daily seasonality, and holiday effects into its additive model. Use this for time series with strong seasonal effects and multiple seasons of historical data.

Seasonal ARIMA: An extension of ARIMA. It supports univariate time series data with a seasonal component.

Exp Smoothing: Assigns greater weighting to more recent observations. This smoothes out time series data using the exponential window function to make predictions.

Simple Exp Smoothing: The simplest of the exponentially smoothing methods. Use this to forecast data with no clear trend or seasonal pattern.

#### Input and output

To use this tool, you must include:

A date column.

### Note

During configuration, you can use the

**Automatic**option where the tool decides the best available date column to use.A value column.

A frequency, which refers to the frequency of the time periods used in the date column.

The results will forecast for that number of periods according to the frequency you select.

You can opt to use the **Automatic** feature in the **Advanced** configuration section. The **Automatic** feature decides the best available settings to use in the following categories:

Trend type

Seasonal type

Seasonal periods

The forecast tool outputs numeric values.

#### For each

Optionally, fill in the **For each** field for more control over how Varicent ELT forecasts your data. When you select a column to analyze by, Varicent ELT might make different predictions for each group.

For example, let's say you have a data set with four columns: seller ID, pay component, payout, date. If you want to forecast a seller's payout, it would be useful to analyze this by pay component. Varicent ELT takes this into account and gives more nuanced results that take pay component over time into consideration. This can also help give better results if the number of rows differs between pay components.

### Note

Due to the computational intensity of building multiple forecasts for each group, by default this feature is limited to splitting the data into 10 groups. If the selection in the **For Each** field would make more than 10 subgroups, the configuration will be ignored.

#### Ignore subgroup limit

Optionally, the **Ignore subgroup limit** option overrides the limit of 10 groups, and allows any number of forecasts to be built.

### Warning

Using this option might cause the pipe to take an extremely long time to build.

#### Multivariate forecasting

Use by adding other columns in your forecasts, instead of using only trends for the output. Toggle on the **Use multiple columns** to activate multivariate forecasting.

Multivariate forecasting uses all forecasters and all train types are applicable.

### Caution

Using this option might cause the pipe to take a long time to build.

The output for multivariate forecasting includes the original columns from your data set and forecasts the data forwards. Bottom rows in the forecast data are based on new data.

#### Charts

You can view the data collected by these forecasting models in the Forecasting Chart, a visual chart representation of your forecasting data.