Vers un modèle de prédiction de la qualité du logiciel pour les systèmes à objets

Modularity has been considered an important software product quality criterion. A software product is considered modular if its components exhibit a high cohesion and its components are weakly coupled. A module has high cohesion if all of its elements are related strongly. In the particular context of object oriented applications, there are different types of coupling between classes. Measuring this types of coupling can helps better understanding the relationship between class coupling and software quality attributes. In this paper, our goal is to validate empirically the hypothesis that some types of coupling have an impact on class error-proneness, using a statistical method and a machine learning technique. MOTS-CLES : mesures de couplage entre classes, modèles prédictifs de qualité, validation statistique, algorithmes d'apprentissage symbolique.