The rapid growth in network demands has led to an increase in network intrusions. A robust network intrusion detection system (NIDS) capable of accurately predicting and identifying intruders is crucial to maintain security. This paper proposes a novel NIDS framework based on ensemble learning and a two-phase hybrid methodology, enhancing both detection accuracy and computational efficiency. Unlike traditional rule-based or signature-based approaches, which struggle with high-dimensional data and evolving cyber threats, our model leverages automated feature selection to handle complex network traffic effectively. The two-phase model combines One-vs-One classifiers with classifiers tailored for specific attack classes, optimizing intrusion detection across varied threats. Additionally, an ensemble approach using Stacking and Voting Classifiers significantly enhances predictive performance, with the Stacking Classifier achieving 100% accuracy in testing. A secure, user-friendly web application built on the Flask framework provides real-time NIDS access to users.