Tamas Fekete

• The transition toward sustainable urban mobility and industrial efficiency requires decision-making tools that go beyond correlation-based analysis to uncover true cause-and-effect relationships. Traditional machine learning models, while effective for prediction, often act as "black boxes," lacking interpretability and failing to reveal the mechanisms underlying complex systems. To address these limitations, this dissertation introduces a modular Causal AI framework for smart decision-making, integrating causal discovery and inference with structured domain knowledge to enhance sustainability outcomes. The framework is validated across three key domains: (1) urban CO2 emissions, (2) shared mobility demand, and (3) SME energy use. The first case study analyzes over 500,000 vehicles to uncover how engine performance and maintenance drive urban emissions. The second study examines shared bike systems, identifying causal impacts of weather patterns, station topology, and temporal demand fluctuations, supporting more adaptive fleet operations. The third applies the framework in a manufacturing SME, identifying the root causes of energy inefficiency and enabling targeted interventions to improve operational performance without compromising productivity. This research advances the interpretability and actionability of AI in sustainability contexts by replacing opaque predictive models with transparent, evidence-based causal reasoning. Algorithms such as PC, FCI, GES, and DirectLiNGAM are employed alongside domain ontologies to uncover valid causal relationships and support decision-making. A hybrid approach also addresses feature selection, dimensionality reduction, and model explainability, making the methodology broadly applicable across diverse sustainability challenges. While the framework demonstrates strong applicability, future work may focus on enhancing real-time scalability, adaptive ontology integration, and broader validation across domains such as electric mobility and smart energy systems. Overall, this thesis contributes a generalizable, interpretable Causal AI framework that enhances systemic understanding and supports sustainable transformation in policy, planning, and industrial decision-making.