Detection of Anomalies in Iot Networks with an Efficientnet-Dicenet Fusion Model

Abstract

With the rapid expansion in the number of devices on the Internet of Things (IoT), it has become quite important to have robust anomaly detection methods to save the IOT networks from growing security vulnerabilities. The EfficientNet-DiCENet fusion model is introduced, which combines the compound scaling of EfficientNet with the depthwise convolutions of DiCENet for further improving anomaly detection in IoT networks. A meta-analysis was used to compare traditional ML and DL methods. The results indicate that the proposed model reaches 98.6% accuracy, with a false positive rate of 3.2%, which exceeds the performance of the existing models in terms of detection performance and computational efficiency. The model also presents low energy consumption (3.7 mJ) to enable real-time, resource-constrained IoT devices. This research contributes to IoT security by presenting an optimized, scalable, and efficient anomaly detection framework. Future work will focus on enhancing adversarial robustness and real-world deployment validation.