Phishing websites continue to pose a significant threat to online security by tricking users into revealing sensitive data such as login credentials and financial information. Traditional detection methods like blacklists typically fail to identify newly launched phishing websites. This paper presents a machine learning-based system that classifies websites as phishing or legitimate based on URL features. Supervised algorithms, including Random Forest and Decision Tree, were used for training and testing. The system extracts lexical and structural features from URLs and uses these to train models. Random Forest outperformed other models in accuracy, robustness, and execution efficiency. A web interface was developed to allow real-time URL submission and classification. Experimental evaluation shows that the system achieves a 96.2% accuracy and a ROC-AUC of 0.98, outperforming baseline blacklist approaches by 17.8% in accuracy. The system is lightweight, making it practical for real-time phishing detection in resource-constrained environments. This work demonstrates the potential of machine learning to enhance phishing prevention and strengthen cybersecurity defenses.