Background: The exponential growth of Internet of Things (IoT) devices has expanded the attack surface for malicious network activities, necessitating robust, low-latency intrusion detection systems.

Objective: This study aims to optimize a Convolutional Neural Network (CNN) architecture to achieve high-accuracy, real-time anomaly detection in IoT network traffic while minimizing computational overhead.

Method: This study uses a simulated dataset created for academic training purposes. A simulated network traffic dataset representing normal and anomalous IoT behaviors was generated. A 1D-CNN model was designed, employing depthwise separable convolutions to reduce parameter count. The model was trained and tested using Python and TensorFlow.

Key Results: The optimized CNN achieved an accuracy of 98.4% and an F1-score of 0.983. Compared to a standard CNN baseline, the optimized model reduced the parameter count by 45% and decreased inference time by 38%, making it highly suitable for edge deployment.