As cyberspace is evolving, new threats in the realm of computer and internet safety are surfacing too. As a result, traditional Intrusion Detection Systems (IDS) are gradually becoming obsolete. Earlier IDS-based security systems focused on predefined signature standards, which made it impossible to detect newly discovered abnormalities and attack variations. Its main issue was the slow rate at which the signature database was refreshed and scaled to keep up with the increasing rate of threat evolution. Researchers continue to employ state-of-the-art methodologies to assure effective threat detection and protection using anomaly-based detection as the task of combating today's cyber threats becomes more difficult. To equip systems with future-proof intrusion detection measures, researchers are adopting sophisticated techniques based on machine learning and deep learning. Because of the variety of modern, complex threats, developing an effective intrusion detection system in a multi-layer attack classification environment is a challenge. Intrusion Detection Systems require high-performance classification algorithms as attackers can readily create intrusive techniques and avoid detection instruments deployed in a computing environment. Furthermore, using a single classifier to effectively detect all types of attacks is complicated. As a consequence, a hybrid approach provides better performance and accuracy. The proposed approach is based on developing a Manifold Approach that is specifically designed to address the aforementioned issue and identify different types of attacks. Each layer of attacks has its own classifier. The proposed approach will be evaluated on the CIC IDS 2018 data set. With the proposed method, the final accuracy obtained is 95.68%, a recall rate of 99.99%, and a better attack detection rate than the baseline classifiers and other existing approaches for different attack categories.