Attribute-value distribution as a strategy for increasing the efficiency of data mining

The paper presents an approach to rule discovery which combines the strategy of targeting a restricted class of rules with a technique for their efficient discovery called attribute-value distribution (AVD). Theoretical and experimental results are presented which show that the reduction in discovery yield, as measured by the proportion of actual rules discovered, is well compensated by the reduction in discovery effort, as measured by the proportion of possible rules examined. Further efficiency gains are possible by parallelisation of the discovery process, since AVD produces a natural decomposition of the discovery task into sub-tasks which can be independently executed on parallel processors