Emerging patterns mining is a data mining task that tries to discover discriminative patterns, which can be used to predict new incoming instances and to describe the behaviour of the underlying data. In the recent years, the description of the dataset has become an interesting field due to the easy acquisition of knowledge by the experts. In this review, we will focus on the descriptive point of view of the task, collecting the existing approaches that have been proposed in the literature and we group them in a taxonomy in order to have a general vision of the task. A complete empirical study demonstrates the suitability of the approaches presented. This review also presents future trends, prospects within emerging patterns mining and benefits of knowledge extracted for emerging patterns.
Results of the statistical test. Selection of the best configuration
Results of the statistical test. Comparison against algorithms
The following table presents the results obtained by the algorithms in the four sets of rules analysed, i.e., the complete set of rules, the minimal rules, the maximal rules and the set of rules whose confidence value in training is greater than a 60 %. You can search for a given algorithm, export the results to several formats and move the position of the columns in order make an easier analysis. For the # of rules and # of varibles quality measures, the results have been tranformated by diving 1/# Rules and 1/# of variables. This is done because it is necessary to minimise the objectives and the range of values for the Friedman test must be normalised. Additionaly. The FPR measures as been transformateb by performing 1 - FPR because we need to minimise the value and the range of values is in [0,1].
Note: A value of zero means the method could not obtain rules for such dataset.
In this section we present the complete experimental study with respect the first part of the study, where the best filter/configuration is selected for each method studied.
First the complete results of the Friedman test for the comparison of each filter on each algorithm and quality measure is presented:
After that, the results of the Shaffer test is presented below. Note that, in this case, the results are presented for all cases, even if the Friedman test shows no significant results.
After the selection of the best configuration is done, a comparison between algorithms for each studied quality measure is made. The results for the Friedman test for each quality measure for each algorithm is:
Due to the results obtained by the Friedman test, it is necessary to perform the Shaffer test. The results obtained for each quality measures are presented below:
Back to top
A graphical 3D representation of Figure 2 in the paper is presented below: