How You Optimize Parameters for Forecasting Methods

This topic explains how you can optimize parameters for forecasting methods in user-defined forecasting profiles that are based on Bayesian machine-learning.

You can optimize forecasting-method parameters for the forecasting profiles that you use in demand, demand and supply, or replenishment plans.

Forecasting Parameters for Optimizing Forecasting-Method Parameters

To optimize parameters for forecasting methods, you need to set values for forecasting parameters, as explained in this table:

Note: If you optimize parameters for forecasting methods using the forecasting parameters listed in this table, then you mustn't use the hyperparameter-tuning functionality.

Forecasting Method	Forecasting Parameter to Be Modified
Modified Ridge Regression (M)	OptimizeRIDGEK When the value is 1, the best value for the RIDGEK method parameter is found during the forecasting process. When the value is 0 (zero), the default value for the method parameter is used. Default Value: 0 Category: Data Smoothing and Cleansing Note: If the value of the CollinearityUseRidge forecasting parameter is 0, the OptimizeRIDGEK forecasting parameter will affect only the Modified Ridge Regression forecasting method. If the value of the CollinearityUseRidge forecasting parameter is 1, the OptimizeRIDGEK forecasting parameter will affect all forecasting methods.
All	OptimizeDimensionReduction When the value is 1, the collinearity and near-collinearity treatment for forecasting methods is optimized during the forecasting process. This forecasting parameter controls the set of causal factors (measures) that are used to avoid issues related to overfitting and collinearity. When the value is 0 (zero), the collinearity and near-collinearity treatment for forecasting methods isn't optimized. Default Value: 0 Category: Data Smoothing and Cleansing

Forecasting Method

Forecasting Parameter to Be Modified

Modified Ridge Regression (M)

OptimizeRIDGEK

When the value is 1, the best value for the RIDGEK method parameter is found during the forecasting process.

When the value is 0 (zero), the default value for the method parameter is used.

Default Value: 0

Category: Data Smoothing and Cleansing

Note: If the value of the CollinearityUseRidge forecasting parameter is 0, the OptimizeRIDGEK forecasting parameter will affect only the Modified Ridge Regression forecasting method. If the value of the CollinearityUseRidge forecasting parameter is 1, the OptimizeRIDGEK forecasting parameter will affect all forecasting methods.

All

OptimizeDimensionReduction

When the value is 1, the collinearity and near-collinearity treatment for forecasting methods is optimized during the forecasting process. This forecasting parameter controls the set of causal factors (measures) that are used to avoid issues related to overfitting and collinearity.

When the value is 0 (zero), the collinearity and near-collinearity treatment for forecasting methods isn't optimized.

Default Value: 0

Category: Data Smoothing and Cleansing

Parameters for Forecasting Methods

This table lists the parameters for optimizing the working of forecasting methods:

Forecasting Method	Parameter
Causal Winters (B)	For optimizing the working of this forecasting method, the parameters should be set as follows: OPTIMIZED ALPHA ITER: 3 (default value) OPTIMIZED BWINT: Yes (default value) OPTIMIZED GAMMA ITER: 10 (default value)
Holt (H)	For optimizing the working of this forecasting method, the parameters should be set as follows: OPTIMIZED ALPHA ITER: 3 (default value) OPTIMIZED GAMMA ITER: 10 (default value) OPTIMIZED HOLT: Yes (default value)

Note: When these forecasting methods are enabled in predefined forecasting profiles, the parameters are optimized by default.