Forecast Strategy
  • 28 Nov 2022
  • 3 Minutes to read
  • Dark
  • PDF

Forecast Strategy

  • Dark
  • PDF


The Forecast Strategy node is used to apply several forecast models to a data set.

Configuration Options

Basic Configuration Options

Setting Description\Parameters
Split Columns The selected column(s) will be used to split the data set over to build additional models.
Target Column The column we are predicting.
Sort By Columns Sort data rows by these columns.
Time Granularity Length of time between records in the input data set. I.e., if each input records represents a week of data, the time granularity is Week.

  • Day
  • Week
  • Month
  • Quarter
  • Year
Role Column Splits the data set into scoring and training data.
Training Value Value in the Role Column for training data.
Scoring Value Value is the Role Column for scoring data.
Predictors The columns we are using to predict.
Model Type Filter Preset selections of models/ensemble models based on the data quality of the input data set.

  • All
  • Good Singal
  • Poor Signal
  • Heavy Zeros
Models See Model Details
Ensemble Models See Model Details
Forecast Horizon Number of periods to forecast forward from the end of the input data set.
Time Column Optional column to seed time values for future forecast periods.
Timeout The specified timeout for the action. If this time is exceeded, the node action will fail, regardless of success of the action.

Model Details

Except where specified, each predictive model comes from the R Fable package. Training data sets are used to build the predictive models, which can then forecast future values.

Some models have been optimized for performance; these are denoted by a Rabbit icon in the model selection dropdown.

Basic Models

Model Description
Mean The forecasts of all future values are equal to the average (or mean) of the historical data.
ARIMA AutoRegressive Integrated Moving Average. Combination of differencing with autoregression and a moving average model. ARIMA models aim to describe the autocorrelations in the data.
Exponential Smoothing State Space Forecasts produced using exponential smoothing methods are weighted averages of past observations, with the weights decaying exponentially as the observations get older; the more recent the observation the higher the associated weight.
Theta Equivalent to simple exponential smoothing with drift.
Linear Regression Time series linear models.
Prophet Prophet is a procedure for forecasting time series data based on an additive model where non-linear trends are fit with yearly, weekly, and daily seasonality, plus holiday effects. It works best with time series that have strong seasonal effects and several seasons of historical data. Prophet is robust to missing data and shifts in the trend, and typically handles outliers well.
Seasonal Decomposition Decomposition combination model using an additive decomposition. Decompose a time series into seasonal, trend and remainder components. Seasonal components are estimated iteratively using STL. Secondary models are used to fit each of the components from the resulting decomposition.
Croston's Based on Croston's method for intermittent demand forecasting; involves using simple exponential smoothing on the non-zero elements of the time series and a separate application of SES to the times between non-zero elements of the time series. Does not produce point forecasts, i.e, a unique forecast for each time point, instead it produces the average demand per future period.
Improved Croston's Based on the above method. The single average demand per future period generated by Croston’s method may be insufficient if predictions for each month for a product that has intermittent demand are desired. Creates a one-step ahead forecast using Croston’s method, then incorporate that prediction back into the training set. Uses this expanded training set to produce another one-step ahead forecast. This process is repeated until you have produced as many predictions as there are time periods in the test set.
Neural Network Feed-forward neural networks with a single hidden layer and lagged inputs for forecasting univariate time series. Does not rely on autoregression to make predictions; can be be utilized for poorer quality/intermittent demand data.

Ensemble Models

Model Description
Good Signal Ensemble Average of Mean, Theta, Exponential Smoothing, ARIMA, Seasonal Descomposition, Prophet, Linear Regression. Output column name shown as GOOD_SIG_ENS.
Poor Signal Ensemble Average of Mean, Neural Network. Output column name shown as POOR_SIG_ENS.
Intermittent Data Ensemble Average of Mean, Improved Croston's, Croston's, Neural Network. Output column name shown as INTER_DATA_ENS.


Action Description
Preview After building any forecast models, the combined result set can be previewed by clicking the Preview button.
Rebuild Model Rebuild any selected models. Produces an .Rdata file that's used to forecast values.

Was this article helpful?