Abstract
Many decisions in fields such as healthcare, finance, energy, or labor market, depend on predictions that ideally are accurate, interpretable, and reliable. In many of those cases, the output being predicted is not a single variable, but it rather has a complex structure, often represented with a vector. Such predictive problems are known by the name of Structured Output Prediction (SOP). We present a predictive framework for SOP based on a conformalized Bayesian decision tree, which exhibits strong predictive performance, inherent interpretability and uncertainty quantification, and reliable predictive intervals obtained through Conformal Prediction (CP). The framework is sector-agnostic and plugs into existing decision processes via simple policy levers like risk levels and abstention rules, which offer several advantages over standard predictive frameworks. As a research-in-progress, we focus on the high-level predictive framework as a decision support tool, its application scenarios, and a multi-faceted evaluation methodology.
Recommended Citation
Andonovikj, Viktor; Mileva Boshkoska, Biljana; and Boškoski, Pavle, "Reliable Decision Support with Conformal Bayesian Decision Tree" (2025). Proceedings of the 2025 Pre-ICIS SIGDSA Symposium. 13.
https://aisel.aisnet.org/sigdsa2025/13