Fuzzy Scenario Analysis

Scenarios enable the user to improve existing fuzzy logic models with their own expert domain knowledge. Rules can be added, removed and edited. The dominance and outcome for rules can also be changed.

To begin making a scenario go to the Analysis module of a fuzzy logic model and start a New Scenario providing it a unique scenario name. This will create a sandbox version of the chosen model.

Once complete, it opens a screen that looks similar to a standard fuzzy logic rules page, with a few extra features:

• Evaluate: After changing a rule, evaluation shows the change from the original model metrics.
• Promote: A scenario can be promoted to become a stand-alone model similar to any other model built from scratch.
• Add rule: Used to create a new rule. 

• Edit: The Edit button under each rule ID allows a rule to be edited which may include any or all of changing, adding, removing antecedents or altering dominance and class.

  The rules that have been added, edited or removed are updated with a relevant status label indicating the change type.

After making changes, a scenario can be evaluated. This will display the new precision, accuracy, and recall on the testing set, and the shift from the original model metrics for each of the output classes.

    Figure 1: Scenario evaluation for a binary classification model.

    Figure 2: Scenario evaluation for a multiclass classification model.

Evaluating a scenario on a regression model will display performance on the relevant metrics.

Figure 3: Scenario evaluation for a regression model.

A scenario can be promoted to a full model by clicking the Promote button. This will then allow all the analysis and deployment options to be made available to your scenario.