<![CDATA[Stacked ensemble models (SEMs) remain widely used for integrating multiple learning algorithms into a single predictive system. However, SEMs continue to face challenges such as accuracy limitations, overfitting, high computational expenses, and limited interpretability. This study conducts a systematic review of 269 peer-reviewed papers published between 2020 and 2025, following the PRISMA (Preferred Reporting Items for Systematic Reviews and Meta-Analyses) methodology to ensure transparency and rigor in article selection. The review identifies key technical issues in SEM implementations and synthesizes their corresponding optimization strategies. To address these challenges, a method-engineering-based modular three-stage framework is proposed, consisting of pre-processing, processing, and post-processing phases. Each stage targets specific weaknesses by improving data quality, optimizing models and hyperparameters, and enhancing interpretability and adaptability. The framework provides a structured foundation that links SEM optimization approaches with their development stages, supporting the design of robust, efficient, and interpretable ensemble models for practical applications.]]>
Copyrights © 2026
