The scientific question in the project concerns machine learning, in particular with regard to the explainability and communicability of recommendations for risk treatment, which are determined or calculated by an artificial intelligence (AI). In this context, the approach considered in the project is a combination of deterministic rules (as in decision trees) with non-rule-based approaches such as regression models. Formally, this involves performing a regression with basis functions generated using fuzzy logic techniques from semantically meaningful defined if-then rules. In other words, the machine learning problem here consists of an optimized selection of if-then rules from a given pool of rules, so that the training data - in this case risk assessments, but alternatively also time series data for the prediction (e.g. via Markov models) of security incidents - are approximated as well as possible (in the sense of metrics or similarity measures to be defined).
KISMS is funded by the FFG Basisprogramme