Linha de pesquisa: Engenharia de Software

Resumo: In the software reliability literature, particularly in the modeling of failure times, probabilistic models based on a single distribution are predominant. However, such models are limited when dealing with the complexity of modern software systems, which are characterized by multiple failure causes and variability in operational profiles. Although widely used during development and testing phases, single-distribution models often perform poorly in real operational contexts due to interactions with the operational environment, which lead to a greater diversity of failure patterns. This study investigates the application of mixture models to software reliability analysis based on real operational data. This approach allows representing intra- and inter-group patterns of time between failures, considering groups as approximations of different operational profiles, as well as the relationship between distinct failure causes. The results indicate that mixture models outperform single-distribution probabilistic models by modeling the variability of failure data in real environments and by accounting for the influence of different operational profiles on failures times.

Link para a defesa: https://teams.microsoft.com/l/meetup-join/19%3aCp6IypkWrnbIq_SCoFbDwE92_...