Enhanced Contractive Porcupine Autoencoder Model With Cognitive Security for Detecting Intrusions in Internet of Things

Since the advent of 5G and the internet of things, security concerns have multiplied rapidly due to the increasing number and sophistication of cyber‐attacks on interconnected devices. Such developments call for the development of advanced intrusion detection systems employing machine learning and deep learning techniques to effectively identify and prevent intrusions. This research introduces the enhanced contractive porcupine multilayer perceptron autoencoder model, aimed at boosting security for 5G and internet of things networks. The proposed model operates through five critical stages: preprocessing, feature scaling, feature extraction, clustering, attack classification, and network recovery. Data normalization is achieved through standardization and label encoding to ensure feature consistency and enhance model performance. Relevant features are extracted using a hybrid spatial–temporal approach. To handle the high volume of traffic data and dynamic distributed denial‐of‐service attacks, co‐adaptive multikernel clustering is employed, allowing for time‐based data clustering and early attack detection. Cluster heads are selected using the binary light spectrum optimizer to manage clusters efficiently. The proposed model accurately classifies various types of attacks and adjusts network topology through control messages to ensure rapid data recovery with minimal delay. The proposed method delivers superior results across several metrics in CICIDS‐2017 dataset: the packet delivery analysis ratio is 90%, throughput analysis ratio is 94%, energy consumption analysis ratio is 96.5%, security analysis ratio is 97.24%, intrusion detection analysis ratio is 98%, and end to end delay is 31.0567 s.