In this respect an important method are the systems for intrusion detection (IDS), whose are designed to identify harmful operation which overrule a functionality of a network. Mobile ad hoc networks (MANETs) (Abraham and Bindu in. 2021 International Conference on Advancements in Electrical, Electronics, Communication, Computing and Automation (ICAECA) (2021)) are groups of wireless devices that can exchange data wirelessly and do not require infrastructure to work. They are hard to secure, due to their decentralized nature and resource shortages. Thus, network intrusion detection system (NIDS) is a very useful tool to ensure network security to detect irregular threats in network traffic. Most of the existing anomaly detection network detection systems are based on using classical machine learning models (e.g. Support vector machine (SVM), etc.). Though some very impressive results have been achieved with these techniques, unfortunately, they are not very robust as they are heavily dependent upon humans engineering transport attributes which is no longer necessary in the age of big data. For achieving detection accuracy of IDS, this research proposes a new multistage optimized fuzzy-based network intrusion detection system (NIDS) architecture that can overcome deficiencies of feature engineering. This research takes a new inside out coverage on the optimal number of training samples needed to achieve the best results by utilizing over sampling methods. It also compares the impact of temporal complexity and detection efficiency on correlation-based and data-gain-based feature selection system, respectively Further, studies are carried out for optimizing data in NIDS in hyper parameter (HP) methods such as genetic algorithm (GA), random search (RS) etc. This is to test the existence of the NSL-KDD dataset on the system. In addition, the detection accuracy levels used to fine-tune hyperparameters increase the accuracy of the model.