The increasing integration of the Internet of Things (IoT) into critical infrastructure has expanded the attack surface for cyber threats, particularly at the network layer. Traditional intrusion detection mechanisms struggle to address the evolving nature of these attacks due to their reliance on predefined signatures and static rule-based approaches. This research proposes a Multi-Layer Perceptron (MLP) based Detection System to enhance the detection and mitigation of network-layer attacks in IoT-driven critical infrastructures. The MLP classifier, a deep learning model, is leveraged for its ability to autonomously learn high-dimensional attack patterns and classify malicious network traffic with superior accuracy. The model undergoes rigorous training and evaluation on benchmark datasets, where it is assessed against traditional machine learning classifiers, including Decision Trees and Random Forest, to validate its performance. Experimental results demonstrate that the MLP classifier significantly outperforms conventional methods in detecting sophisticated network-layer threats, achieving high accuracy and minimal false-positive rates. The study highlights the scalability, adaptability, and real-time threat detection capabilities of MLP, reinforcing its strategic importance in modern cybersecurity frameworks. The findings contribute to advancing deep learning-driven intrusion detection systems, offering a robust and proactive approach to securing IoT-enabled critical infrastructure against evolving cyber threats.