Datenbestand vom 13. Juni 2019
Tel: 089 / 66060798
Mo - Fr, 9 - 12 Uhr
Fax: 089 / 66060799
DER VERLAG IST IN DER ZEIT VOM 12.06.2019 BIS 23.06.2019 AUSCHLIESSLICH PER EMAIL ERREICHBAR.
aktualisiert am 13. Juni 2019
978-3-8439-0062-1, Reihe Informatik
On the principle of heterogeneous redundancy based Bayesian approach to integrate static and dynamic fault prediction models
137 Seiten, Dissertation Technische Universität Kaiserslautern (2011), Hardcover, B5
The comparison of static and dynamic models clearly demonstrates that the two modeling classes are significantly different in their modeling approach and especially in their deficiencies. The existing model selection and combination approaches are restricted to either the static or the dynamic modeling methodology. Unfortunately, these approaches suffer from the same deficiencies and result in similar limitations and sources of failure. This thesis combines static and dynamic models to reduce the impact of each methodology’s specific deficiencies. It weights the models according to their prediction accuracy given certain data and according to their methodology specific conditions. One challenge of the combination of static and dynamic models is that they yield different output values. Static models predict faults and dynamic models predict failures. Therefore, this thesis proposes a method that enables the combination of these different predictions. This thesis improves reliability prediction through the combination of static and dynamic models. Compared to the state of the art practice in software fault and failure prediction an approach to combine both methodologies has been defined. In order to execute the approach, a system has been derived to combine static and dynamic predictions, two combination methods have been defined for different software development phases, criteria have been determined to apply the correct combination approach in each phase, the mathematical foundation for the weighted combination has been established, and a model has been developed to integrate expert and environmental information into the combination. This hybrid combination approach is evaluated using a tool prototype on data of a simulation study. Therefore, the predictions of the developed approach have been compared to the predictions of the conventional models. It is shown that the developed approach yields more stable and more credible predictions than the conventional single models.