Applying and Testing Mult-iClass and Multi-Output Algorithms in the Mapping of Security Requirements with Technologies and Best Practices

• Autor(a) principal: Batista, Pedro Miguel Marques
• Data de Publicação: 2022
• Tipo de documento: Dissertação
• Idioma: eng
• Título da fonte: Repositório Científico de Acesso Aberto de Portugal (Repositórios Cientìficos)
• Texto Completo: http://hdl.handle.net/10400.6/13115
• topic: Classificação Multiclasse; Algoritmos Multioutput; Aprendizagem Automática; Cibersegurança; Classificação Multilabel; Estrutura de Segurança; Internet das Coisas; Segurança por Desenho; Domínio/Área Científica::Engenharia e Tecnologia::Engenharia Eletrotécnica, Eletrónica e Informática

Nowadays, companies are increasingly deploying more Internet of Things (IoT) devices into the market without considering the security requirements of these systems. Platforms like the SECURIoTESIGN framework attempt to minimize the number of devices that are released with these vulnerabilities, by informing and guiding interested developers about possible secure implementations without needing to contact a security expert (even though it does not replace his knowledge). The modules use questionnaires, whose answers generate a set of recommendations, obtained using heuristics or fixed rules. To facilitate the development of this modules and the provided recommendations, as well as minimizing the need of a security expert, embedding of Machine Learning (ML) was proposed and pursued, being applied to two modules of the platform. In this dissertation the research needed to implement ML in this context is explored and explained, along with the implementation details on both modules, including the creation of a dataset containing all possible answer combinations, automatically. Furthermore, an analysis of the generated dataset was made, how to artificially augment it, and its usage examined, using different variations of available data, for training and testing various multi­class and multi­output models, therefore allowing to simulate situations where resources could not be obtained or an expert was not available. It was possible to conclude that the usage of multi­class and multi­output algorithms presented positive results, when performed with different variations of training data, allowing to conclude that implementing ML in this context may bring advantages to the platform.