A Modified Linknet Architecture for Attack Detection in Internet of Things

As the IoT continues to proliferate across various domains, ensuring the security of IoT networks against malicious attacks remains a critical challenge. This work proposes a novel method to enhance the attack detection in IoT environments by employing a Modified LinkNet (MLinkNet) architecture. Addressing inherent challenges in existing methodologies, the SMOTE is employed to mitigate the class imbalance issue in IoT attack datasets, there by improving the model performance. The Modified Link Net model deviates from conventional encoder-decoder framework is introduced by integrating the fused pooling layers and an improved activation function designed to preserve high-resolution features critical for accurate attack identification. Rigorous experimentation on IoT attack datasets demonstrates superior detection accuracy and computational efficiency compared to traditional methods. This research contributes a robust solution to IoT security challenges, advancing the field with a scalable and effective model capable of adapting to emerging cyber threats in IoT ecosystems.