Elie Prudhomme

Association Rule Interestingness: Measure and Statistical Validation

The search for interesting Boolean association rules is an important topic in knowledge discovery in databases. The set of admissible rules for the selected support and condence thresholds can easily be extracted by algorithms based on support and condence, such as Apriori. However, they may produce a large number of rules, many of them are uninteresting. One has to resolve a two-tier problem: choosing the measures best suited to the problem at hand, then validating the interesting rules against the selected measures. First, the usual measures suggested in the literature will be reviewed and criteria to appreciate the qualities of these measures will be proposed. Statistical validation of the most interesting rules requests performing a large number of tests. Thus, controlling for false discoveries (type I errors) is of prime importance. An original bootstrap-based validation method is proposed which controls, for a given level, the number of false discoveries. The interest of this method for the selection of interesting association rules will be illustrated by several examples.

Statistical inference and data mining: false discoveries control

Data Mining is characterized by its ability at processing large amounts of data. Among those are the data “features”- variables or association rules that can be derived from them. Selecting the most interesting features is a classical data mining problem. That selection requires a large number of tests from which arise a number of false discoveries. An original non parametric control method is proposed in this paper. A new criterion, UAFWER, defined as the risk of exceeding a pre-set number of false discoveries, is controlled by BS FD, a bootstrap based algorithm that can be used on one- or two-sided problems. The usefulness of the procedure is illustrated by the selection of differentially interesting association rules on genetic data.

Maps ensemble for semi-supervised learning of large high dimensional datasets

In many practical cases, only few labels are available on the data. Algorithms must then take advantage of the unlabeled data to ensure an efficient learning. This type of learning is called semi-supervised learning (SSL). In this article, we propose a methodology adapted to both the representation and the prediction of large datasets in that situation. For that purpose, groups of non-correlated attributes are created in order to overcome problems related to high dimensional spaces. An ensemble is then set up to learn each group with a self-organizing map (SOM). Beside the prediction, these maps also aim at providing a relevant representation of the data which could be used in semi-supervised learning. Finally, the prediction is achieved by a vote of the different maps. Experimentations are performed both in supervised and semi-supervised learning. They show the relevance of this approach.