Rumor detection on social media has gained significant attention in recent years, with machine learning (ML) playing a key role in automating this task. This study evaluates the performance of several ML models on three benchmark datasets—GossipCop, Fake News Challenge (FNC-1), and Politifact—sourced from Kaggle. These datasets offer diverse examples of true and false information. The analysis reveals that the Gradient Boosting Classifier is the most robust and adaptable model across datasets, while Support Vector Machine and Logistic Regression show strengths in precision and specific detection tasks. The study highlights existing research gaps and concludes with insights into the current state of the art in online rumor detection.