Analysis of various interestingness measures in classification rule mining for traffic prediction

Recently, an evolutionary algorithm named Genetic Network Programming with Estimation of Distribution Algorithm (GNP-EDA) has been proposed and applied to extract classification rules for solving traffic prediction problems. The measures such as the support, confidence and χ² value are adopted to evaluate the interestingness of a large number of rules extracted from traffic databases in the above data mining method. In data mining, many other measures have been proposed to evaluate the interestingness of association patterns. These measures usually provide different and conflicting results. Many studies investigate that the effects of different measures depend on the concrete applications. We rarely know what measures are the appropriate ones for the traffic prediction application. Therefore, a novel approach to select the right measure for the classification rule mining has been proposed in this paper. The simulation results show that the proposed interestingness measure selection approach is a powerful tool to select the right measure for the traffic prediction application, leading to the increase of the classification accuracy.