Advanced intrusion detection systems are required due to the quick uptake of cloud computing and the growing complexity of cyber threats, especially Denial of Service and Distributed Denial of Service attacks. Deep learning architectures are becoming more popular because traditional IDS techniques frequently falter in dynamic, large-scale settings. Using datasets including CICIDS2017, NSL-KDD, and UNSW-NB15, this paper assesses the effectiveness of well-known DL architectures for intrusion detection, including Convolutional Neural Network, Recurrent Neural Networks, Long Short-Term Memory, and others. Key performance indicators such as accuracy, precision, and false positive rates are examined to compare the efficacy of these models. The findings show that some designs, like ResNet and Self-Organizing Map, perform well in structured environments but poorly on complicated datasets like KDDTest-21. Another important data gap highlighting the need for more research in this area is that most models do not automatically adjust to unexpected threats. This work aids in the creation of intelligent, scalable systems for changing network environments by evaluating the efficacy of DL-based IDS solutions.