Enhancing Sentiment Classification with Stacked Ensembles and Differential Evolution: An Empirical Study on Ukraine Conflict Tweets

Abstract Background/Objectives: Social media platforms generate massive volumes of sentiment-rich data during conflicts, but their unstructured and noisy nature poses serious challenges for accurate sentiment analysis. This study aims to enhance the classification of sentiment in tweets related to the Ukraine conflict, addressing issues of class imbalance, sarcasm, and rapidly evolving narratives. Method: A stacked ensemble framework optimized using Differential Evolution (DE) is proposed, combining both traditional machine learning classifiers and deep learning models within a meta-learning architecture. Novel improvements to the DE algorithm include feature-aware mutation, adaptive crossover, sentiment-weighted fitness (macro-F1), and dynamic parameter adaptation, all designed to boost model robustness and generalization. The approach was empirically validated on a curated dataset of 42,000 Ukraine-conflict tweets. Findings: Baseline ensemble models achieved approximately 82% accuracy, while the DE-optimized ensemble reached 91.7% accuracy, with corresponding improvements in precision, recall, and F1-score. The system achieved a ROC–AUC of 0.93, with Neutral sentiment recall increasing by 6%. Statistical significance was confirmed through paired t-tests and Wilcoxon signed-rank tests, verifying the superiority of the optimized model. The framework also demonstrated scalability and low inference latency, indicating suitability for real-time crisis dashboards and misinformation detection. Novelty: This work introduces a DE-enhanced ensemble architecture that unifies feature-aware optimization and sentiment-specific adaptation within a single framework—an advancement not commonly seen in conflict-related sentiment analysis. The integration of DE with meta-learning for social media data provides a robust, generalizable, and computationally efficient solution. Keywords: Sentiment Analysis, Stacked Ensemble, Differential Evolution, Social Media, Ukraine Conflict, Optimization