rule-based systems i: decision tables, decision trees...

MSc Business Information Systems

Rule-Based Systems I: Decision Tables, Decision Trees, Visual Rule Modelling

Knut Hinkelmann

Prof. Dr. Knut Hinkelmann 2MSc BIS/

Rule-based Systems

Rule-based Systems consist of Rule base Fact base Inference engine

Rules are of the form IF <antecedent> THEN <consequent>

KnowledgeBase

InferenceComponent

Facts

Rules


Different Types of Rule-based Systems

Condition-Action Rules (Production Rules) <consequent> is a set of actions <antecedent> is a set of conditions Example:

IF car X has been sold THEN delete X from the catalogue

Logical rules <antecedent> and <consequent> are formulas of first-order

logic IF … THEN is equivalent to implication Example:

IF X is a man THEN X is mortal


Representation of Rules

There is a large variety of rule representations:

decision tables

decision trees

graphical representation

textual representation logical formulas Structured English „Code“


Decision Tables

A decision table is a compact form to represent a whole set of rules

A decision table can represent condition-action rules and also logical rules Condition-Action rules:

Logical Rules: The effectsrepresent possible decisionvalues

The second column represents a set of rules: one column for eachcombination of possible values for condition

Action entriesActions

Condition alternativesConditions

Effect entriesEffects

Condition alternativesConditions


Decision Table for Printer Diagnosiseach columnrepresents one rule

This decision table represents condition-action-rules


Decision Table for Health Insurance

Effects

N N N N N N Y Y Y Y Y Y 1. Deductible met? L L H H D D L L H H D D 2. Type of visitN Y N Y N Y N Y N Y N Y 3. Participating Physician?

X1. Reimburse 50%X 2. Reimburse 70%

X3. Reimburse 80%

X X X X 6. Impossible or N/A XXX X 5. No reimbursement

X 4. Reimburse 90%

12 11 10 9 8 7 6 5 4 3 2 1 Conditions

http://web.sxu.edu/rogers/sys/decision_tables.html

Reimbursement depends on whether decuctible is already met, whether the patient visited thedoctor‘s office (D), a hospital (H) or a lab (L) and whether the Doctor is a Participating Physician


Decision Tables – Reducing Combinations If effects for a several combinations are the same, the

combinations can be combined, reducing the numer of rules

Example: If Cause 1 is „Y“ and Cause 2 is „N“, the the effect does not

depend on the values of Cause 3 In this case, the value of Cause 3 does not need to be taken

into account

Causes Values 1 2 3 4 5 6 7Cause 1 Y, N Y Y Y N N N NCause 2 Y, N Y Y N Y Y N NCause 3 Y, N Y N - Y N Y NEffectsEffect 1 X XEffect 2 X X X

Combinations

X XX

X

Causes Values 1 2 3 4 5 6 7 8Cause 1 Y, N Y Y Y Y N N N NCause 2 Y, N Y Y N N Y Y N NCause 3 Y, N Y N - - Y N Y NEffectsEffect 1 X X XEffect 2 X X X

Combinations

Y N

X XX

X


Decision Trees

Decision trees are a graphical representation of rules Each inner node corresponds to a decision Each edge represents an alternative value for the decision The leaf nodes represent actions or effects

Deduciblemet?

no reimbursement

Type ofvisit

ParticipatingPhysician?

N

YL

H

DY

N 50%

90%

80%

70%

MSc Business Information Systems

Visual Rule Modelling


VisiRule – a graphical modelling tool

VisiRule™ is an extension to WIN-PROLOG from Logic Programming Associates Ltd.

VisiRule allows experts to build decision models using a graphical paradigm

VisiRule allow to graphically represent forward chaining rules with access to Prolog

VisiRule models can be interpreted and executed

VisiRule models can be exported to other programs and integrated into existing web and desktop standards


Layers of VisiRule

VisiRule creates charts (layer 1)

VisiRule generates Flex code (LPA ‘s Expert System Product) (layer 2)

Flex code in turn generates Prolog (layer 3)

The underlying Prolog allows to do almost anything, including call C functions using a built-in predicate called winapi/4

VisiRule

Flex

WIN-PROLOG

Integration with VB, Delphi, ODBC


Answer the question … and get result


A simple visible chart

Start node

Question

Expressions

End nodes


Question Types


Question Types

Single Choice This is the default option. The menu produced will onlyallow the user to select one of the items on the menu.

Multiple Choice This allows the user to select any or none of the items on the menu.

Number Input Instead of a menu, this option provides an input box intowhich the user can enter any number

Integer Input This is like Number Input, but only allows the user to enteran integer.

Set Input An input box is also provided by this option. The user cantype in a list of items, separated by a space character. For example: red amber green.

Name Input Another input box is provided into which the user can typea word or phrase.


Generating and running executable code

To run a ruleset:• Right click in background• Select „Show Code …“• Click on [Run]


Generating and running executable code

Running a ruleset:• Questions are asked one after the other• Select the appropriate answer• The result is shown at the end


Representing a Decision in VisiRuleVisiRule diagrams are graphical representations of forward chaining rules:


The same rule set can also be represented as a decision table:


Travel Advisory represented as Decision Table and Decision Tree


Dividing a Decision Tree into Subtrees usingContinuation Boxes


Dividing a Decision Tree into Subtrees usingContinuation Boxes (Cont.)


Statement Box

The function of a statement box is to calculate a value frominformation that is already known.

Statement boxes have three elements: an editable name (balance_plus_order in example below) an editable local variable (X in example below) a statement assigning a value to the variable using the operator

„is“ (X is balance + order_total.)(Note: On the right of „is“ there is editable Prolog code which isused to calculate the value


Statement Box with an Arithmetic Expression


Example: Calculating Leap Years

rule-based systems i: decision tables, decision trees...

Documents