Linguistic-Summary-Based Feature Selection for Qualitative Decision Support

Abstract

Qualitative decision models such as DEX rely on a hierarchical aggregation of elementary criteria into interpretable decision rules. While this structure supports transparency and explainability, it also makes model construction sensitive to the number and interdependence of elementary criteria. This paper proposes a decision-aware feature selection approach that supports parsimonious DEX modelling by explicitly accounting for directional relevance and redundancy among candidate criteria. The method employs linguistic summaries to estimate asymmetric dependencies enabling features to be ranked by their directional influence on the decision concept and pruned when mutually redundant. Applied to internal migration modelling in Serbia, the approach reduces a highly interdependent socio-economic feature set to a smaller number of interpretable and policy-relevant criteria, despite weak correlations with the target variable. The results demonstrate that linguistic-summary-based directional analysis can uncover meaningful regularities not captured by symmetric measures and provide a bridge between numeric data analysis and qualitative decision support modelling.

Sandro Radovanović

Assistant Professor at University of Belgrade

My research interests include machine learning, development and design of decision support systems, decision theory, and fairness and justice concepts in algorithmic decision making.