<p>The most popular technique for protecting networks in the face of growing cyberattacks is a network intrusion detection system (NIDS). There are several network intrusion detection systems with high accuracy available. However, because these systems are not entirely trained for detecting low-occurring attacks, these NIDS occasionally fail to identify attacks with low occurrence but great impact. Therefore, automatic anomaly detection systems are necessary to defend the network from all kinds of attacks, regardless of how frequent they are. Therefore, an efficient and optimal machine learning-based intrusion detection system is proposed, which can detect low occurring threats in addition to the most frequent ones. In this paper we propose an optimized intrusion detection system employing a weighted approach and SMOTE balancing that helps the IDS to detect such kinds of attacks whose frequency is low but impact is high. These approaches are applied on different ML models to get more accurate and efficient NIDS. To make the system efficient and accurate, the proposed model also used PSO to optimize the features. The process determines which features are important based on centrality in its connected components and finds Pareto-dominant pairings of informative and associated features. PSO gives a more concise and useful representation of the dataset. Experimental results show that the proposed model only requires 40% of the features available yet produces a more accurate model. Results show that the detection rate of fewer attacks is approximately the same in proposed Model A, but in proposed Model B it significantly increases from 22.61 to 39.68%. All Experimental results are performed on the UNSW-NB15 Dataset.</p>