Predictive Analytics for Unveiling Patterns in Online News Popularity

Source abstract

The online news industry is highly competitive, driven by readers' demand for immediate access to information. Predicting article popularity before publication can provide strategic advantages for content optimization. This study utilizes the Online News Popularity dataset from the UCI Machine Learning Repository, where articles with more than 2000 shares are classified as popular. Four classification models namely logistic regression, k-nearest neighbours (KNN), random forest, and naive Bayes were evaluated to compare their predictive performance. Results show similar performance across all models, with the KNN model (k=200) achieving the highest accuracy (0.644) and the largest area under the curve (AUC=0.584). The findings demonstrate the feasibility of machine learning approaches for forecasting news popularity prior to publication. Such predictive capability enables media organizations to optimize content strategies, allocate resources efficiently, and enhance competitiveness in the rapidly evolving digital news environment.