Cette dissertation présente un ensemble d'algorithmes visant à en permettre un usage rapide, robuste et automatique des « Support Vector Machines » (SVM) non supervisés dans un contexte d'analyse de données. Les SVM non supervisés se déclinent sous deux types algorithmes prometteurs, le « Support Vector Clustering » (SVC) et le « Support Vector Domain Description » (SVDD), offrant respectivement une solution à deux problèmes importants en analyse de données, soit la recherche de groupements homogènes (« clustering »), ainsi que la reconnaissance d'éléments atypiques (« novelty/abnomaly detection ») à partir d'un ensemble de données. Cette recherche propose des solutions concrètes à trois limitations fondamentales inhérentes à ces deux algorithmes, notamment I) l'absence d'algorithme d'optimisation efficace permettant d'exécuter la phase d'entrainement des SVDD et SVC sur des ensembles de données volumineux dans un délai acceptable, 2) le manque d'efficacité et de robustesse des algorithmes existants de partitionnement des données pour SVC, ainsi que 3) l'absence de stratégies de sélection automatique des hyperparamètres pour SVDD et SVC contrôlant la complexité et la tolérance au bruit des modèles générés. La résolution individuelle des trois limitations mentionnées précédemment constitue les trois axes principaux de cette thèse doctorale, chacun faisant l'objet d'un article scientifique proposant des stratégies et algorithmes permettant un usage rapide, robuste et exempt de paramètres d'entrée des SVDD et SVC sur des ensembles de données arbitraires.

When systems designers are making decisions about which components or technologies to select for a design, they often use experience or intuition to select one technology over another. Additionally, developers of new technologies rarely provide more information about their inventions than discrete data points attained in testing, usually in a laboratory. This makes it difficult for system designers to select newer technologies in favor of proven ones. They lack the knowledge about these new technologies to consider them equally with existing technologies. Prior research suggests that set-based design representations can be useful for facilitating collaboration among engineers in a design project, both within and across organizational boundaries. However, existing set-based methods are limited in terms of how the sets are constructed and in terms of the representational capability of the sets. The goal of this research is to introduce and demonstrate new, more general set-based design methods that are effective for characterizing and comparing competing technologies in a utility-based decision framework. To demonstrate the new methods and compare their relative strengths and weaknesses, different technologies for a power plant condenser are compared. The capabilities of different condenser technologies are characterized in terms of sets defined over the space of common condenser attributes (cross sectional area, heat exchange effectiveness, pressure drop, etc.). It is shown that systems designers can use the resulting sets to explore the space of possible condenser designs quickly and effectively. It is expected that this technique will be a useful tool for system designers to evaluate new technologies and compare them to existing ones, while also encouraging the use of new technologies by providing a more accurate representation of their capabilities. I compare four representational methods by measuring the solution accuracy (compared to a more comprehensive optimization procedure's solution), computation time, and scalability (how a model changes with different data sizes). My results demonstrate that a support vector domain description-based method provides the best combination of these traits for this example. When combined with recent research on reducing its computation time, this method becomes even more favorable.