An enhanced possibilistic fuzzy linear regression model using conditional non-symmetric fuzzy numbers

This paper presents an enhanced Possibilistic Fuzzy Linear Regression (PPFLR) model for datasets characterized by asymmetric and imprecise information. Existing fuzzy regression methods generally rely on symmetric triangular or trapezoidal fuzzy coefficients, which often produce wide prediction intervals and weak interpretability. The proposed model employs conditional non-symmetric pentagonal, hexagonal, and octagonal fuzzy numbers that allow independent control of left and right spreads. The estimation problem is formulated as a linear programming model that minimizes total fuzziness while ensuring that all observed values lie within the predicted fuzzy bounds at specified confidence levels. A real used-vehicle pricing dataset is analyzed, and preliminary diagnostics including VIF, Shapiro–Wilk, and Durbin-Watson tests confirm the suitability of the classical regression baseline. The PPFLR model produces narrower spreads and more stable predictions than classical possibilistic regression and crisp least-squares estimation. Based on the development of new fuzzy structures and confidence-based constraints, approximately 70% of the proposed framework represents methodological novelty beyond existing approaches. The results demonstrate the practical applicability of non-symmetric fuzzy numbers in valuation and decision-making problems involving incomplete or vague information.