Using Subgroup Discovery to Analyze the UK Traffic Data

Rule learning is typically used in solving classification and prediction tasks. However, learning of classification rules can be adapted also to subgroup discovery. Such an adaptation has already been done for the CN2 rule learning algorithm. In previous work this new algorithm, called CN2-SD, has been described in detail and applied to the well known UCI data sets. This paper summarizes the modifications needed for the adaptation of the CN2 rule learner to subgroup discovery and presents its application to a real-life data set the UK traffic data confirming its appropriateness for subgroup discovery in real-life applications through experimental comparison with the CN2 rule learning algorithm as well as through the evaluation of an expert. Furthermore we make the first step towards the comparison of the new CN2-SD algorithm to another state-of-the-art subgroup discovery algorithm SubgroupMiner by applying both algorithms to a slightly different data set the UK traffic challenge data set. The results of this application are presented in the form of ROC curves, showing CN2-SD’s potential in finding descriptions (subgroups) for minority classes, while SubgroupMiner found ‘better’ subgroups when trying to describe the majority class given the problem at hand.