Validation levels and standards depending on models types and functions - Institut Polytechnique de Paris Accéder directement au contenu
Communication Dans Un Congrès Année : 2019

Validation levels and standards depending on models types and functions

Résumé

The concept of model validation is central to modeling but remains loosely defined and strongly dependent on context. We propose in this contribution an overview of what is considered as validation, depending on model type or function. We first review existing definition and criteria for a broad range of model types, including predictive models, statistical models, analytical models, simulation models. We develop in the last case the role of sensitivity analysis methods, and illustrate the development of new validation methods in the case of geosimulation models with spatial sensitivity analysis techniques. We then consider broad model function families introduced by (Varenne, 2017), and develop what model validation means for each class (which are observation, understanding, theory construction, communication, and decision-making) with specific cases for some subclasses. This typology of model validation depending on model function appears to be complementary to the approach by model type. We conclude by contextualizing modeling within a knowledge framework for complex systems (Raimbault, 2017), and how validation can be seen as an intermediate between different knowledge domains in the iterative process of the construction of a perspective.
CCS2019_SatelliteSimexplo_Raimbault.pdf (723.4 Ko) Télécharger le fichier
Origine : Fichiers produits par l'(les) auteur(s)

Dates et versions

halshs-02441891 , version 1 (16-01-2020)

Identifiants

  • HAL Id : halshs-02441891 , version 1

Citer

Juste Raimbault. Validation levels and standards depending on models types and functions. Conference on Complex Systems 2019 - SIMEXPLO II Satellite, ISC-PIF, Jan 2020, Singapore, Singapore. ⟨halshs-02441891⟩
58 Consultations
21 Téléchargements

Partager

Gmail Facebook X LinkedIn More