Parameters tab

The Data Processing Parameters section contains parameters used to prepare and validate input data for AI Engine processing.

You can modify this information:

Zero fill sales History

Indicates whether missing sales history data points are replaced with zero values.

Note: By default, this switch is set to On.

Extend zero fill from start horizon

Indicates whether zero filling is extended from the start of the historical horizon to the first sales data point.

Note: You can modify this field only if the Zero fill sales History switch is set to On.

Extend zero fill until global last sale

Indicates whether zero filling is extended until the last available global sales data point.

Note: You can modify this field only if the Zero fill sales History switch is set to On.

Handle missing data in attributes

Indicates whether missing values in input attributes are processed.

Note: By default, this switch is set to Off.

Impute threshold

The percentage threshold used to determine when missing attribute values are imputed.

Note: 

• By default, the value in this field is set to 25%.
• You can modify the value in this field only if the Handle missing data in attributes switch is set to On.

Keep attributes

The attributes retained during missing data processing.

Note: The values displayed in the list are populated from the Engine Input tab.

Drop threshold

The percentage threshold above which attributes with missing values are excluded from processing.

Note: 

• By default, the value in this field is set to 70%.
• You can modify the value in this field only if the Handle missing data in attributes switch is set to On.

Attribute selection

Indicates whether the attribute selection configuration is enabled.

Note: By default, this switch is set to On.

Position

The position at which attributes selection is performed. Possible values:

• Pre-processing - faster, sample-based
• Post processing - more accurate
• Auto - decides based on data

Note: 

• By default, the value in this field is set to Post-processing.
• You can modify the value in this field only if the Attribute selection switch is set to On.

Max selected

The maximum number of attributes that can be selected during the attribute selection process.

Note: 

• By default, the value in this field is set to 15.
• You can modify the value in this field only if the Attribute selection switch is set to On.
• The following two selection methods are supported:
  • random_bar: A fast, noise-aware method that keeps only the attributes proven to carry more predictive signal than chance, so weak or spurious drivers are filtered out automatically.
  • multi_criteria: A more thorough method that evaluates each attribute from several complementary angles and retains the set where predictive value clearly drops off, giving a well-balanced and stable selection.

Enforced attributes

The attributes that must be included in the selection, regardless of the selection method.

Note: 

• The values in the list are populated from the Engine Input tab.
• You can modify the value in this field only if the Attribute selection switch is set to On.

Selection method

The method used to select attributes. Possible values:

• random_bar
• multi_criteria

Note: 

• By default, the value in this field is set to random_bar.
• You can modify the value in this field only if the Attribute selection switch is set to On.
• The following two selection methods are supported:
  • random_bar: A fast, noise-aware method that keeps only the attributes proven to carry more predictive signal than chance, so weak or spurious drivers are filtered out automatically.
  • multi_criteria: A more thorough method that evaluates each attribute from several complementary angles and retains the set where predictive value clearly drops off, giving a well-balanced and stable selection.

Outlier detection

Indicates whether the outlier detection is performed on the input data.

Note: By default, this switch is set to Off.

Outlier detection method

The method used to identify outliers. Possible values:

• Custom
• Naïve

Note: 

• By default, the value in this field is set to Custom.
• You can modify the value in this field only if the Outlier detection switch is set to On.
• Two outlier detection methods are supported:
  • Naive: detects outliers by comparing each value to the entire series distribution (global view). We use it for relatively stable data (no strong trends/seasonality). It requires the definition of test threshold parameter to control how far a point must deviate from the global series to be flagged (typical values: 4–6).
  • Custom: detects outliers by comparing each value to a local neighborhood (window) around it (context-aware). We use it when data has trends, seasonality, or regime changes where local context matters.

Statistical test

The statistical test used during the outlier detection. Possible values:

• Tau - correlation-based
• Step - sudden jump detection

Note: 

• By default, the value in this field is set to Tau.
• You can modify the value in this field only if:
  • The Outlier detection switch is set to On.
  • The value in the Outlier detection method field is set to Custom.

Test threshold

The threshold used during the outlier detection.

Note: 

• By default, the value in this field is set to 5.
• You can modify the value in this field only if the Outlier detection switch is set to On.

Window position

The relative position of the analysis window used during the outlier detection. Possible values:

• Left - history
• Right - future
• Center - balanced

Note: 

• By default, the value in this field is set to Left.
• You can modify the value in this field only if:
  • The Outlier detection switch is set to On.
  • The value in the Outlier detection method field is set to Custom.

Window size

The size of the analysis window used during the outlier detection.

Note: 

• By default, the value in this field is set to 100.
• You can modify the value in this field only if:
  • The Outlier detection switch is set to On.
  • The value in the Outlier detection method field is set to Custom.

Use statistical assortment

Indicates whether statistical assortment is calculated based on statistical patterns. The Statistical Assortment feature helps decide which products to include in forecasts based on their sales activity. This feature detects statistically the introduction and discontinuation date for a planning entity based on historical sales activity. When enabled, the statistical introduction and discontinuation dates are calculated and reported.

Note: By default, this switch is set to Off.

Historical assortment

Indicates whether the statistical calculation of historical assortment is performed.

Note: 

• This switch is enabled only if the Use statistical assortment switch is set to On.
• When enabled, the statistical introduction and discontinuation dates are applied. The engine does not zero fill any date outside the statistical introduction / discontinuation dates interval. The engine does not learn on any date outside this interval.

Future assortment

Indicates whether the statistical calculation of future assortment is performed.

Note: 

• This switch is enabled only if the Use statistical assortment switch is set to On.
• When enabled, the engine excludes the planning entities that are statistically discontinued from forecasted.