A New Method for Ranking the Discovered Rules Obtained from Data Mining Using Data Envelopment Analysis

Data mining techniques, i.e. extraction of patterns from large databases, are extensively used in business. Many rules may be obtained by these techniques and only a few of them may be considered for implementation due to limitation of budgets and resources. Evaluating and ranking attractiveness and usefulness of the association rules are of paramount importance in data mining. In the earlier studies carried out on identifying mentally interesting association rules, most methods required writing information or asking users for explicit differentiation of interesting rules from uninteresting ones. These methods involve detailed calculations and they may even lead to inconsistent conclusions. To solve these problems, this article proposes the application of the double frontiers Data Envelopment Analysis (DEA) Approach for selecting the most effective association rule. In this approach, in addition to the best relative efficiency of each association rule, its worst relative efficiency is considered. Comparing with the traditional DEA, double frontiers DEA Approach is capable of identifying the most efficient association rule correctly and easily. As an advantage, the proposed approach is more efficient than the earlier works in this concern, as far as calculations are concerned. Applicability of our DEA-based method for measuring efficiency of association rules will be shown by multiple criteria using an example of market basket analysis.