Each innovation brings with it a fresh set of problems to solve. The information stored on a system should be protected against access by anybody who has not been permitted to do so. Regarding system security, the first and most important task is to install and maintain a reliable and effective network intrusion detection system. If a structure has an interruption detection procedure in place, it may be able to estimate the total number of interruptions that will occur in the future and the present. The use of artificial intelligence approaches for organization-based interruption locations has been around for more than two decades already, and there are a variety of ways available. An effective interruption detection system will probably continue to be a topic of debate for a considerable amount of time. Companies must create better methods for keeping their systems and information secure from gatecrashers and hackers as the number of digital assaults and the bulk of system evidence continue to grow at an alarming pace. Due to the integration of more sophisticated security apparatuses into cutting-edge project designs, the volume of security events and ready information created continues to expand, making it more difficult to trace down the perpetrators of the assault as well as the gatecrashers. When it comes to managing the detection, reaction, and organization of security incidents and possible assaults on their systems, organizations are forced to depend on new ways to support and supplement human investigators for the first time. In particular, the emphasis of this Thesis is on differentiating between normal system traffic information and dangerous system traffic data. This study's objectives are to enhance the characteristics of generating system information using particle swarm optimization (PSO), and then to create a Network Intrusion Detection System using directed learning using a completely related Deep Neural Net (DNN) using directed learning (NIDS). It is feasible to develop sophisticated neural system models with the use of the NSL-KDD dataset that surpasses the constraints of the KDD Cup2009 interruption recognition datasets, which have been regularly utilized in the past. In experiments using the NSL-KDD datasets, it has been shown that deep neural networks with molecular swarm augmentation are very effective in terms of accuracy and recognition rates.