This document discusses developing a machine learning system to classify URLs as malicious or benign in real-time. The system was trained on data collected from tweets during two large events - the Super Bowl and Cricket World Cup. A multi-layer perceptron (MLP) model achieved the best performance, correctly classifying 72% of URLs from the unseen Cricket World Cup data within 30 seconds. The Bayesian model performed best in early stages, achieving 66% accuracy within the first 60 seconds. Analysis of the MLP model revealed that bytes received and remote IP address were important indicators of malicious URLs.