Decision Mining Trees

The Decision Mining tool in the Viabl Platform can be used to generate probabilistic Decision Trees (or Regression Trees for numeric Outcomes) from historic data records. These Trees can then be combined with other Viabl Platform knowledge Objects, to build complex decision automation applications that combine new knowledge derived from data combined with knowledge from human experts.

To use Decision Mining requires the prior setting up of an Viabl Platform Decision mining Server Data Mining Server

Once a Decision Mining Server has been set up, you can create a new Decision Mining analysis by double-clicking the New Decision Mining Tree Tool (under AI Tools)

• Select an existing Data Table (or click Data Mining Server Management icon to add a new Data Table)
• Enter optional SQL WHERE Clause to filter in the required rows)
• Exclude any unwanted fields
• Select the Outcome field

Building a Decision Tree from Data

There are 2 types of Mined Trees Decision (Classification) Trees with discrete Outcomes and Numeric (Regression) Trees with numeric outcomes.

The Tool bar icons provide the following operation:

Fullt dataset information

Induce a full Tree

Access Monitored Data Sets

Calculate tree classification accuracy

Data Mining Settings

Mined Decision Tree Settings (Discrete outcomes)

Mined Decision Tree Settings (Numeric outcomes)

Fuzzified Tree Inference

Decision Mining trees with Numeric outcome also support Fuzzy inference which can potentially increase the accuracy of prediction when Numeric Attributes are utilised in the Mined Tree.

Executing Decision Mined Trees & Accessing Outcome Statistics

Decision Mined tree Objects are executed just like a standard Viabl tree. i.e. they be set as the start object, used (drag-dropped) inside other decision trees, or using the Object's .run method in a Scripts.

Once the execution of a decision mined tree is complete, it's outcome is assigned

• For List outcome trees, the outcome value with the highest probability in the fired Leaf node, or a fixed value (see "Data Mining Settings > Outcome Rate Threshold")
• For Numeric outcome trees, the outcome is the average value of the fired Leaf node.

There are a number of special properties which can be used (in a JavaScript Script) to access the results of executing the data mining tree:

The the inferred outcome for the fired leaf (for List Decision Mined Trees) e.g. If the data mined tree is called Decision then we can display:

The value of the inferred outcome:

xpertrule.message(#Decision.val());

The probability of the inferred outcome:

xpertrule.message(#Decision.prop("prob"));

We can also access the probability of a given outcome value of the fired Leaf node (for List Decision Mined Trees) e.g. If the data mined tree is called Decision we can display the probability of the 2nd outcome value:

xpertrule.message(#Decision.value(2).prob);

*The average and standard deviation of the fired leaf (for Numeric Decision Mined Trees)) e.g. If the data mined tree is called Balance we can display:

The average numeric outcome of the fired leaf:

xpertrule.message(#Balance.val());

The standard deviation of the fired leaf:

xpertrule.message(#Balance.prop("stddev"));