A meta-possibilistic fuzzy framework for robust bivariate correlation estimation
Applied Soft Computing, cilt.200, 2026 (SCI-Expanded, Scopus)
- Yayın Türü: Makale / Tam Makale
- Cilt numarası: 200
- Basım Tarihi: 2026
- Doi Numarası: 10.1016/j.asoc.2026.115411
- Dergi Adı: Applied Soft Computing
- Derginin Tarandığı İndeksler: Science Citation Index Expanded (SCI-EXPANDED), Scopus, Applied Science & Technology Source, Compendex, INSPEC
- Anahtar Kelimeler: Bivariate correlation, Ensemble learning, Meta fuzzy functions, Outlier, Robust estimation
- Karadeniz Teknik Üniversitesi Adresli: Evet
Özet
Traditional correlation measures, such as Pearson's coefficient, are widely used for assessing bivariate associations; however, their reliability may deteriorate under non-normal distributions, nonlinear dependencies, and the presence of outliers. Although several robust estimators have been proposed, selecting a single estimator that performs consistently well across different data conditions remains challenging. To address this limitation, this study proposes an ensemble-based dependency estimation framework called the Meta Possibilistic Fuzzy Correlation Function (MPFCF). MPFCF integrates multiple correlation estimators and dynamically weights their contributions using Possibilistic Fuzzy c-Means clustering, supported by a novel cluster validity measure, the Penalized Fuzzy Validity Index. The framework was evaluated through extensive simulation experiments and real-world benchmark datasets, including the Hawkins–Bradu–Kass synthetic dataset. The results show that MPFCF consistently produces lower symmetric mean absolute percentage error (SMAPE) values and more stable estimates compared to individual estimators. An ablation study confirmed the structural robustness of the framework, with the mean rank of MPFCF remaining within a narrow range (1.167–1.667) despite substantial changes in the estimator pool. Furthermore, the generalizability of the framework was validated by extending the estimator pool to include modern [0,1] dependency measures such as Chatterjee's ξ, Robust Distance Correlation, Baak's ∅K, and HSIC, where MPFCF maintained its superior global rank of 1.717. Statistical validation using the Friedman test and Nemenyi post-hoc analysis confirmed that these improvements are statistically significant. The proposed MPFCF provides a robust, adaptive solution for dependency estimation in complex real-world data analysis tasks involving nonlinear relationships and outliers.