Akademik Veri Yönetim Sistemi
Applied Soft Computing, cilt.190, 2026 (SCI-Expanded, Scopus)
Selecting a single regression model can be unreliable when data contain complex structures or correlated features. Although model combination is often proposed as a remedy, it remains unclear when this approach is most effective and how different models should be combined. This study addresses these issues and introduces Meta Fuzzy Regression Functions (MFRFs) as a refined model combination framework. MFRFs integrate predictions from multiple regression models through fuzzy membership based weighting. Models produced after feature selection procedures are included in this process, and when features are correlated or redundant, these models tend to perform more stably. The fuzzy weighting mechanism naturally assigns higher weights to such models, allowing the meta function to adaptively emphasize the most informative predictors. Experiments on six simulated and four real-world datasets show that MFRFs mostly outperform individual models. The approach reduces test prediction errors and achieves meaningful improvements in Mean Absolute Percentage Error (MAPE). These findings demonstrate that MFRFs offer an effective and interpretable solution to model selection uncertainty and enhance regression performance across diverse data settings. The results on several datasets show that the proposed method improves the average MAPE by approximately 25% compared to individual models and by 41% compared to the second-best ensemble learning method.