Model-Driven, Logic-Supported Exploratory Dependability Analysis

Analyzing empirical dependability data now exceeds domain experts' capabilities without adequate machine support. This is due to the complexity of modern IT systems, their many interdependent features, and the extensive time series involved. This paper presents a model-driven approach to Exploratory Data Analysis (EDA), which is guided by formal logic reasoning over an abstract data model. This supports domain experts in formulating and validating hypotheses, as well as guiding the exploration process. Observations are transformed into qualitative abstractions, facilitating Answer Set Programming (ASP) based formal reasoning over system behavior and dependencies. Counterexample-guided inductive learning (CEGIL) and Rough Set Theory (RST) are used for iterative hypothesis refinement and uncertainty management. The integrated framework enables precise, transparent, and systematic dependability assessment using commonsense engineering principles, thereby assisting analysts in the evaluation process with pertinent expert insights.