Association rules are a simple but very useful form of data mining that describe the probabilistic co-occurrence of certain events within a database. They were originally designed to analyse market-basket data, in which the likelihood of items being purchased together within the same transaction were analysed.
In the ARMADA tool (Association Rule Mining And Deduction Analysis) the rules are presented in the form of;
Item1, ItemN => ItemX
The rule is evaluated using two numerical measures.
SUPPORT: The number of occurrences, within the data set, of all components of an association rule occurring within a single transaction (a transaction can be thought of as a row in a conventional database).
CONFIDENCE: This is the probability that, given the antecedent (left hand side of the rule) being true, the consequent (right hand side) is true.
ARMADA contains several visualisation techniques for analysing quantative and qualative measures of each rule, as well as several strategies for undertaking data mining which can be used as exclusive or combinatorial methods.
To run GUI version type ARMADA in command line.
NOTE: A major bug fix was applied to version 1.4 which affected newer versions of MATLAB not generating large rules. This is now resolved.
Homepage:
http://www.ebi.ac.uk/~malone |