<p>Intrusion detection has become indispensable for modern cybersecurity as evolving cyberattacks continue to threaten networked environments. The rapid escalation of sophisticated cyberattacks has created an urgent need for intelligent intrusion detection approaches capable of identifying threats with high accuracy and low latency. In this paper, we propose a new method to detect the cyberattacks in network traffic by utilizing a three-tier Deep Learning (DL) approach and hybrid optimization algorithm. The proposed three-level DL framework consists of Convolutional Neural Networks (CNNs) for spatial feature extraction, Support Vector Machines (SVMs) for the reliable classification, and Siamese Recurrent Neural Networks (SRNNs) for pairwise similarity learning to perform detection of known and zero-day attacks. Using GPSA (Golden Pelican Search Algorithm) as a hybrid optimization model to select features and tune hyperparameters further improves performance. It combines the exploratory characteristics of the Golden Jackal Optimization (GJO) with the exploitation abilities of the Pelican Optimization Algorithm (POA). Experimental results on the standard cybersecurity datasets show that our model achieves better accuracy, precision, sensitivity and specificity in comparison to existing algorithms. This paper has presented the potential of combining multi-tier DL with bio-inspired hybrid optimization for next-generation network systems.</p>