Rule-based machine learning algorithm

May / 07

 

Rule includes at least one "precondition" and one "postcondition".

 

*precondition evaluates a data point

*postcondition giving hypothesis output fot the data point

ex) if a fruit is red and circular, it is an apple.

precondition : "red and circular" 

postcondition: "it is an apple"

rule:  if a fruit is red and circular, it is an apple.

 

There are two rules which are Classification rule, and Association rule

 -> Classification rule: classify a data point to a typical class.

 -> Association rule: find a relationships between data points.

 

1 Rule

• 1.1 Conditional relation ship: if a precondition of rule a retuns "true" for a typical data point, the data point will do postcondition of the rule.
• 1.2 Sequential evaluation of rules: if a data returns "true" and be applied postcondition, the data will not be evaluated from next condition ( there are multiple conditions in a model)
  • 1.2.1 set of rules evaluate a data point one by one. 
  • 1.2.2 if a data point is returns "false" from rule A, the data point will be evaluated by rule B.
• 1.3 Variable: it is possible to a condition includes a variable
  ex) if a fruit is "X", it is "Y" : X can be red, yellow, green ...., Y can be apple, banana, melon.

 

 

2. Sequential covering (rule-based machine learning algorithm): it is one of the algorithms that use classification rule

• 1. the model create a rule that includes as much data as the postcondition returns true.
• 2. apply the rule to the dataset
• 3. classify the data points based on the rule
• 4. except the data points that were classified in step 3
• 5. return to 1 till all data points are classified.

https://rmit.instructure.com/courses/135554/pages/w7-learning-materials-2?module_item_id=5930139

3. Apriori algorighm (rule-based machine learning algorithm): it is one of the algorithms that use association rule. Normally it is used in market baket anlaysis.

ex) to find the relationships between the items that are sold by consumers

• Key factors)
  • Support : the percentage of a certain relationship
    ex) the dataset represnts the set of items purchased by customers in a specific market. If a relation is found which is that normally when customer purchase bread, butter is included in the set of items. 
    
    The support of relation between bread and butter
    = (num of purchase that includes both bread and butter) / total num of puchase
  • Confidence: if A is existing in a data point, the percentage of B is also included in the data point.
    ex) if a puchase includes bread, the percentage of butter is also included in the purchase.
    confidence = (the num of purchase includes both bread and butter) / (total num of purchase)
    
    
  • Lift: it is key metric used to evaluate the strength of an association rule. if the lift value is greater than 1, it is considred that there is a positive association between the two items.
    ex) finds association between bread and butter.
    Lift = confidence of bread and butter / (the num of purchase which includes butter / num of total purchase)
• Aprior algorithm operation
  1. find most frequent item
  2. find the one more item have a relationship with the item (set of item)
  3. if the support of relationship(association) is greater than minimum support values that the user decide, expand the set of items.
     
     * The mean of 1,2,3 is create a rule* => create rule and make it increasingly more specific hypotheses