ABSTRACT This study proposes an integrated framework to analyze inferred multidimensional poverty‐exit potential in Tunisia using nationally representative MICS 2021 data. Because the data are cross‐sectional, the framework does not observe true household transitions over time. Instead, it constructs a continuous Fuzzy Poverty Exit Index (FPEI) that captures proxy evidence of partial progress away from multidimensional deprivation. The index is combined with a leakage‐controlled interpretation layer: XGBoost models predict high FPEI status, while SHAP attributions are computed using predictors that explicitly exclude all indicators used to construct the FPEI. To explore policy‐relevant heterogeneity, we estimate conditional effect heterogeneity with causal forests under standard unconfoundedness and overlap assumptions, interpreting the results as observational conditional estimates rather than definitive causal effects. We also derive a fuzzy clustering typology of exit‐potential profiles and provide a descriptive spatial diagnostic of regional disparities across Tunisian governorates. Results show that education, housing quality, asset accumulation, and territorial location are the strongest correlates of higher exit potential, with substantial heterogeneity across households and regions. Three profiles—resilient, vulnerable, and structurally supported exit‐potential groups—summarize the diversity and stability of observed conditions. The framework contributes to multidimensional poverty measurement by combining fuzzy scoring, explainable prediction, heterogeneity analysis, and spatial diagnostics, offering practical guidance for geographically targeted and multisectoral anti‐poverty policy.
Belhadj et al. (Mon,) studied this question.