Pierre-Emmanuel Jouve

A filter feature selection method for clustering

High dimensional data is a challenge for the KDD community. Feature Selection (FS) is an efficient preprocessing step for dimensionality reduction thanks to the removal of redundant and/or noisy features. Few and mostly recent FS methods have been proposed for clustering. Furthermore, most of them are ”wrapper” methods that require the use of clustering algorithms for evaluating the selected features subsets. Due to this reliance on clustering algorithms that often require parameters settings (such as number of clusters), and due to the lack of a consensual suitable criterion to evaluate clustering quality in different subspaces, the wrapper approach cannot be considered as a universal way to perform FS within the clustering framework. Thus, we propose and evaluate in this paper a ”filter” FS method. This approach is consequently completely independent of any clustering algorithm. It is based upon the use of two specific indices that allow to assess the adequacy between two sets of features. As these indices exhibit very specific and interesting properties as far as their computational cost is concerned (they just require one dataset scan), the proposed method can be considered as an effective method not only from the point of view of the results quality but also from the execution time point of view.

Optimisation and evaluation of random forests for imbalanced datasets

This paper deals with an optimization of Random Forests which aims at: adapting the concept of forest for learning imbalanced data as well as taking into account users wishes as far as recall and precision rates are concerned. We propose to adapt Random Forest on two levels. First of all, during the forest creation thanks to the use of asymmetric entropy measure associated to specific leaf class assignation rules. Then, during the voting step, by using an alternative strategy to the classical majority voting strategy. The automation of this second step requires a specific methodology for results quality assessment. This methodology allows the user to define his wishes concerning (1) recall and precision rates for each class of the concept to learn, and, (2) the importance he wants to confer to each one of those classes. Finally, results of experimental evaluations are presented.

Dynamic Control of the Browsing-Exploitation Ratio for Iterative Optimisations

A new iterative optimisation method based on an evolutionary strategy. The algorithm is proposed, which proceeds on a binary search space, combines a Genetic Algorithm and a knowledge extraction engine that is used to monitor the optimisation process. In addition of the boosted convergence to the optima in a fixed number of iterations, this method enables to generate knowledge as association rules. The new algorithm is applied for the first time for the design of heterogeneous solids in the frame of a combinatorial program of catalysts development.

A method for aggregating partitions, applications in K.D.D.

K.D.D. (Knowledge Discovery in Databases) methods and methodologies nearly all imply the retrieval of one or several structures of a data set. In practice, using those methods may give rise to a bunch of problems (excessive computing time, parameters settings, ...). We show in this paper that some of these problems can be solved via the construction of a global structure starting from a set of sub-structures. We thus propose a method for aggregating a set of partial structures into a global one and then present how this method can be used for solving several traditional practical problems of K.D.D.