Rule Learning

Rule LearningIt is a rule that can be used to distinguish unseen examples. It is usually obtained by learning from a training data set. It is usually an IF-THEN rule, a type of unsupervised learning, and is often classified as a type of classification.

Rules in rule learning

Rules: They have clear semantics and can describe the objective laws or domain concepts implied by data distribution.

There are two types of rules for rule learning:

• Propositional Rule

A simple declarative sentence consisting of "atomic propositions" and logical connectives "and, or, not, and".

Examples:

Rule 1: (viviparous = no) ∧ (can fly = yes) → birds

Rule2: (viviparous = no) ∧ (living in water = yes) → fish

• First-order Rule

Unlike propositional rules that only deal with simple declarative propositions, first-order logic additionally includes assertions and quantification. First-order rules can express complex relationships and are also called relational rules.

Methods for generating rules

• Direct Method: Induce rules directly from the training set;
• Indirect Method: converted from the decision tree.

The purpose of rule learning

The goal of rule learning is to generate a rule set that can cover as many examples as possible. Sequential covering is a commonly used approach. Before a new rule is learned in the training set, the training examples of the rule are covered, and the remaining training examples form the training set and the above process is repeated.

Since only a portion of the data is processed at a time, this approach is also called a divide and conquer strategy.