Discovering association rules partially devoid of dissociation by weighted confidence

Useful rule formation from frequent itemsets in a large database is a crucial task in Association Rule mining. Traditionally association rules refer to associations between two frequent sets and measured by their confidence. This approach basically concentrates on positive associations and thereby do not study the effect of dissociation and null transactions in association. Though the effect of dissociation has been studied in association, the impact of null transactions has been generally ignored. Some scholars have identified both positive and negative rules and thus studied the impact of the null transactions. However there is no uniform treatment towards inclusion of null transactions in either positive or negative category. We have tried to bridge these gaps. We have established a uniform approach to mine association rules by combining the effect of all kinds of transactions in the rules without categorizing them as positives and negatives. We have proposed to identify the frequent sets by weighted support in lieu of support and measure rules by weighted confidence in lieu of confidence for useful positive rule generation taking care of the negativity through dissociation and Null Transaction Impact Factor. We have shown that the weighted support, weighted confidence approach increases the chance of discovering rules which are less dissociated compared to the traditional support-confidence framework provided we maintain same level of minsupp and minconf in both cases.