A1626
Title: From prediction to understanding: Causal discovery for data science and AI applications
Authors: Shohei Shimizu - The University of Osaka (Japan) [presenting]
Abstract: Causal discovery aims to go beyond prediction and infer cause-effect relationships directly from data, which is essential for reliable decision-making in science, industry, and government. Recent advances in causal discovery are presented together with real-world applications in medicine and environmental science, in a way accessible to a broad data science and AI audience. On the methodological side, extensions of existing approaches to more realistic settings are introduced, including cases with hidden common causes and datasets that mix discrete and continuous variables. These challenges are common in practice but remain difficult for standard methods. Results show how causal discovery can still provide useful insights under such conditions. On the application side, these methods are demonstrated on real problems. In medicine, causal discovery helps identify clinically meaningful relationships from observational data, supporting more interpretable analyses. In environmental science, a study on adverse outcome pathways is presented, where causal relationships among biological events are inferred from data. Overall, causal discovery complements machine learning by enabling a shift from prediction to understanding how interventions influence outcomes.
