Personal tools
Laboratoire d'InfoRmatique en Image et Systèmes d'information

Skip to content. | Skip to navigation

Laboratoire d'InfoRmatique en Image et Systèmes d'information
UMR 5205 CNRS / INSA Lyon / Université Claude Bernard Lyon 1 / Université Lumière Lyon 2 / École Centrale de Lyon
You are here: Home > membres

Léopold Ghemmogne Fossi


PhD student

Team Distributed Systems, Information Retrieval and Mobility
Institution Institut National des Sciences Appliquées de Lyon
Location Blaise Pascal (INSA)
E-mail leopold.ghemmogne-fossi at
Contact details Publications Thesis
Subject A game theoretic approach to Feature selection for efficient Multi-Criteria Decision Making: two classification use cases
Abstract Many decision support systems rely on the implementation of a set of "event-action" or "event-condition-action" rules.
The aggregation of these rules is most often done by simple disjunction (aggregator "or"). The objective of this thesis is to adapt the available methodologies to optimize the performance of classifiers based on a set of rules; improve the management of the set of rules over time (addition, deletion, modification, etc.) in order to adapt to the temporal evolution of the processed data flow.
One of the fields of application of our work concerns the detection of fraud on credit cards in the framework of the collaboration with ATOS Worldline, partner of INSA Lyon. In particular, we propose to:
1) Develop methods to make improvements over the state of the art, in terms of: overall performance of classifiers ("recall", "precision")
2) Also improve in the management of the rules considering this time the performance of a rule, evaluated not in isolation,but in terms of its contribution within the set of rules. Regarding the evaluation of the individual contribution of a rule within the set of rules, since the aggregation is not additional, we will use "Power Index" concepts: concepts borrowed from the domain of Theory of Collaborative Games, such as Shapley Value. The other field of application will be the classification of tweets: we aim through the "Power Index": to better define the coverage of an event (defined by tweets) by selecting an optimal set of keywords.
Advisor Lionel Brunie
Advisor Elod Egyed-Zsigmond

Last update : 2018-01-16 14:07:55