Cross Attention with Deep Convolutional Recurrent Autoencoder Network Intrusion Detection System in Internet of Things

Internet of Things (IoT) technology demands highly robust security measures across various domains.This need has become increasingly critical due to the inherent computational resource limitations and production constraints in IoT devices.Traditional Intrusion Detection System (IDS) methods face challenges when detecting unknown or diverse attacks as they struggle to interpret network traffic and effectively classify it.Additionally, these devices often find it difficult to capture relevant network features.This research proposes a Cross Attention with Deep Convolutional Recurrent Autoencoder Network (CA-DCRAN) for efficiently capturing temporal dependencies in IoT data and handling high dimensional data using autoencoder mechanism.Initially, pre-processing is performed for normalization and standardization of feature vectors, alongside classifying them using a label encoder.Feature selection and dimensionality deduction are performed using Periodic Boundary -Mantis Search Algorithm (PB-MSA), which offers superior accuracy.Classification using CA-DCRAN efficiently identifies attacks combined with convolutional layers while reconstructing network traffic patterns.The CA-DCRAN achieves a commendable accuracy of 99.70% on IoTID20, 99.86% on N-BaIoT and 98.95% on NSL-KDD datasets, thereby outperforming the existing Self-Attention -Deep Convolutional Neural Network (SA-DCNN).