Skip to Main content Skip to Navigation
New interface
Journal articles

Calibrer les comportements d'agents à partir de données réelles

Philippe Mathieu 1 Sébastien Picault 1 
1 SMAC - Systèmes Multi-Agents et Comportements
CRIStAL - Centre de Recherche en Informatique, Signal et Automatique de Lille - UMR 9189
Abstract : The new application domains of multiagent-based simulation provide very large databases, which must be take into account not only for validating simulation results, but also for calibrating the behaviors of the agents. Indeed, the confidence in simulation predictions and explanations highly depends on the statistical realism of the agents. In this paper, we propose a method for automatically retrieving behavioral prototypes from statistical measures. This method has been experimented within the context of consumer behavior. The agents are endowed with the same overall behavior, but are given different profiles based on the data analysis. They are put into a spatially realistic store, where their purchase reproduce the original clusters. The knowledge retrieving process is general enough to be used in various application domains. Thus, we argue that such techniques are crucial to enhance the predictive accuracy of multi-agent simulations and make them a powerful decision support tool.
Complete list of metadata

Cited literature [38 references]  Display  Hide  Download
Contributor : CRIStAL Equipe SMAC Connect in order to contact the contributor
Submitted on : Saturday, August 5, 2017 - 5:44:37 PM
Last modification on : Tuesday, November 22, 2022 - 2:26:15 PM


Files produced by the author(s)



Philippe Mathieu, Sébastien Picault. Calibrer les comportements d'agents à partir de données réelles. Revue des Sciences et Technologies de l'Information - Série RIA : Revue d'Intelligence Artificielle, 2014, 28 (4), pp.463-484. ⟨10.3166/ria.28.463-484⟩. ⟨hal-01071977⟩



Record views


Files downloads