A Feature-Driven Machine Learning Approach For Netflix Content Prediction and Caching

Understanding content popularity is critical for streaming platforms like Netflix to optimize user engagement and predict future trends. This paper explores feature analysis for analyzing Netflix content popularity, focusing primarily on hours viewed as a core metric. Using machine learning models, such as XGBoost, RandomForest, and hybrid approaches, this study examines the interplay of content attributes, viewer preferences, and temporal trends. In addition, a topic modeling approach based on series plots is implemented to identify latent themes and patterns that influence content popularity over time. The proposed framework integrates data preprocessing, model training, hybrid methodologies, and thematic analysis to reveal deeper insights into engagement dynamics. The outcomes demonstrate the effectiveness of selecting the right features with simpler models like Logistic Regression to obtain an accuracy of 82% on the Netflix engagement dataset and provide actionable strategies for enhancing recommendation systems, tailoring content delivery, and aligning offerings to meet evolving viewer preferences.