Question

IF–THEN rules can be used for classification. What are some performance measures that can be used to evaluate the accuracy of the rule, and how do these measures differ? What are the advantages and limitations of each performance measure? You should compare at least two performance measures for evaluating the performance of an IF–THEN rule.

Answer 1

IF THEN rules can be used for classification :-

• The IF THEN classification is rule based where the model is represented as a set of IF THEN rules.
• Example to consider rule1
• if age = youth and person = yes
• then buy_laptop = yes
• If the condition is true for given example then the precondition is satisfied.
• The IF part of the rule is known as precondition.And the THEN part is known as rule consequent. If the result is converted from decision tree to classification rules. then it can be as follows.
• Mutually exclusive rules -- If the rules is independent from each other then it is mutually exclusive rule. ever record is covered by at most one rule.
• Exhaustive rules -- In this rule each and every record is covered by at least one rule.
• Use default class to make the rule set

Some performance measures that can be used to evaluate the accuracy of the rule and difference of these measures :-

• The rule based classification of IF THEN include -- Rule induction algorithm, Rule ranking measures and Class prediction algorithm.
• Rule induction algorithm -- It is process of extracting IF THEN from the data which are directly done using sequential covering algorithm.
• Rule ranking measures -- This represent some values which is used to measure usefullness of the rule and provide accurate prediction.
• The using rule ranking measures improve efficiency and speed and are used in class prediction algorithm and give ranking to the rules.

Advantage --

• It is highly expressive as decision trees
• It is easy to generate.
• The performance are comparible to decision trees.
• It can classify new instance rapidly.

Limiation --

It reduces rules with very small coverage

The next rule can be identical to previous rule.

Underestimating of accuracy of rules.