Research Project Mining Negative Rules in Large Databases
Research Project Mining Negative Rules in Large Databases using GRD
Rule Discovery Paradigms • Classification Rule Discovery • Association Rule Discovery • Generalized Rule Discovery • Rule: A => B – A is the antecedent – B is the consequent
Classification Rule Discovery • Aim: Make predictions from rules discovered in data • Discover a small number of rules that cover most of the training data • Focus on a single consequent
Association Rule Discovery • Aim: Searches database to find strong associations between itemsets • Itemsets are subsets of the dataset • Coverset: set of transactions that an itemset (A) occurs in
Association Rule Discovery (Contd. ) • Support of A => B: coverset (A U B) / |D| • Confidence of A => B : coverset (A U B) / • coverset (A) Min. constraints are defined to accept a rule. – Minimum support (frequent itemsets) – Minimum confidence (interest)
Generalized Rule Discovery • Uses the concepts of Association Rule Discovery • Uses the search method from Classification Rule Discovery – The OPUS Algorithm for an unordered Search. • User specifies alternate minimum constraints
Aims Find negative correlations between Itemsets in a database. This will be achieved by extending the GRD technique • Rule: A => ~B, ~A => ~B
AIMS (Contd. ) tidsets A = diffset ~A • With very little additional computational overheads the negative associations can be calculated • Assess whether the results of negative correlations are potentially interesting or not
- Slides: 9