摘要 |
PURPOSE: A data mining method and a recording medium capable of being read by a computer are provided to supply an inductive rule and a sequential pattern with respect to an association between different attributes. CONSTITUTION: RULES(set of initial single-condition rule) and BEST(the most important rules out of the RULES) are initialized as an empty set, respectively(S701). A single-condition rule is created with respect to all values of all attributes and inserted in the RULES(S702). The 'k' numbers having highest 'H'(significance of each rule) value in the RULES are inserted into the BEST(S703). While the RULES is not an empty set(S704), a loop is progressed. If the RULES becomes an empty set, contents of the REST are output(S705). If the RULES is not an empty set, one rule is selected in the RULES(S707). If a success probability is '1'(S708), the rule is deleted(S709). If a success probability is not '1', a range of 'H' value of a subdivided rule is calculated(S710). If a range of the calculated 'H' value is less than "H*"(the lowest value out of the 'H' value) value(S711), the 'H' value is deleted(S709). While a condition attribute to be appended newly exists, a subdivided rule is created and inserted in the RULES. If the new rule is greater than "H*", the new rule is exchanged with a rule having "H*" in the BEST and the "H*" value is updated.
|