Forecasting

With Time Series Forecasting, you can calculate forecast data into the future based on the time series data frame from the past.

Input Data

Input data should be a time series data frame. It should have following columns.

• A date/time column with Date or POSIXct data.

• A value column with numeric values that you want to forecast into the future.

How to Access This Feature

• Click "+" button and mouse over "Run Analytics ...", and select "Run Time Series Forecast" submenu to open "Run Time Series Forecast" dialog.

Add forecast data to time series data frame

After "Run Time Series Forecast" dialog is opened, follow the steps below to calculate forecast data.

1. Select date/time column with "Date / Time Column" dropdown. This columns is the time axis along which forecasting is performed.

2. Select aggregation level from "Aggregation Level" dropdown. Input data is aggregated with this time unit. For example, if you select "Day" and there are multiple rows that falls within a same day, the values for those rows are aggregated to form single row for the date. As a result, this becomes the time unit for the resulting time series data frame. Aggregation level can be one of the following.

• Day

• Week

• Month

• Quarter

• Year

1. Select value column from "Value Column" dropdown. This is the value of interest that you want to forecast.

2. Select aggregation function from "Aggregation Function" dropdown. This is how the multiple rows that falls under same time period (the one you specified with "Aggregation Level".) is aggregated. Its value can be one of the following.

• sum

• mean

• count

• median

• max

• min

• first

• last

• mad - Median Absolute Deviation

• sd - Standard Deviation

• var - Variance

1. Specify the number of time period for which you want to forecast the value of interest in "Forecasting Time Period" text field.

2. (Optional) Specify additional parameters in "Parameters" section.

• Capacity (Cap) Value or Data Frame - Possibly achievable maximum value. Can be numbers like market size, or some kind of theoretical limit.

• Strength of Seasonality

• Return Yearly Seasonality - Take yearly seasonality into account while forecasting.

• Return Weekly Seasonality - Take weekly seasonality into account while forecasting.

• Number of Potential Changepoints

• Flexibility of Automatic Changepoint Selection

• Potential Changepoints

• Strength of Holiday Effect

• Holiday Column - A categorical column that indicates type of holiday for the day of the observation. When logical column is selected, TRUE means it is a holiday and FALSE means it is not.

• Data Frame for Holidays - Holiday definition data frame
• Use Built-in Country Holiday Data - By setting this option to Yes, you can specify list of country codes whose holiday data should be considered in making forecast.

• Country for Holiday Data - Specify list of country codes here. For supported countries, take a look at this table.
• MCMC Samples for Full Bayesian Inference

• Width of Uncertainty Intervals

• Number of Simulations for Uncertainty Intervals

1. Click "Run" button.

Output Data

The output data comes with the following in addition to the input data.

• Rows for forecasted data with future date values

• Additional Columns about forecasted values

  • forecasted_value - The forecasted value.

  • forecasted_value_high - Upper bound of forecasted value

  • forecasted_value_low - Lower bound of forecasted value

  • trend

  • trend_high

  • trend_low

  • seasonal

  • seasonal_low

  • seasonal_high

  • yearly

  • yearly_high

  • yearly_low

  • weekly

  • weekly_high

  • weekly_low

  • cap_model

  • cap_forecast

Here is an example of line chart created from the output data. DEP_DELAY (blue line) is the input value column. Note that orange line for the forecasted_value goes into future beyond the end of the blue line.

Reference

• An Introduction to Time Series Forecasting with Prophet Package in Exploratory