Cyber-security is an emerging technology that involves securing digital information using methods such as firewalls, antivirus software, and intrusion detection systems (IDS). However, intrusion detection system (IDS) datasets often have high dimensionality and imbalance, leading to challenges in model performance such as biased classification, extremely heavy model weight, increasing redundancy, complexity, and resource utilization. To overcome the challenges above, in this paper, a lightweight IDS is designed utilizing a PSO-based technique. It incorporates a novel fitness function that considers the imbalanced nature of data and applies an ensemble of different machine and deep learning-based detection models. Several base and ensemble models are used to evaluate the detection performance of the proposed approach in terms of accuracy, precision, recall, f1-score, false alarm rate (FAR), and prediction time. The proposed model's accuracy is 88.48% and 87.87% on UNSW-NB15 and NSL-KDD datasets, respectively, using the XGBoost detection model.