Predictive Modeling in Communication Studies: A Systematic Review of Feature Selection Techniques for Data-Driven Insights

Abstract

The increasing adoption of machine learning techniques in communication research has brought new opportunities for predictive modeling, audience analysis, and media content evaluation. However, the challenges of ensuring both accuracy and interpretability remain central to advancing computational communication studies. This review synthesizes recent literature on hybrid feature selection frameworks that integrate statistical, heuristic, and machine learning-based approaches. We examine how these methods improve prediction performance while maintaining transparency and theoretical relevance in communication contexts. Key applications include predicting audience engagement, analyzing social media discourse, modeling media effects, and exploring communication networks. The review also highlights current limitations, such as overfitting risks, data bias, and insufficient theoretical integration, and suggests directions for future research that bridge computational methods with core communication theories. By critically evaluating hybrid feature selection strategies, this study provides communication scholars with insights into balancing methodological rigor with interpretability in machine learning applications.