Intelligent Intrusion Detection in IoT Networks Using Hybrid LSTM2LP Algorithm

Abstract

Intrusion identification has become essential for blocking unwanted traffic in IoT networks. To protect networks and information from various cyber-attacks, a wide range of intrusion detection techniques have been developed. Machine learning, capable of handling massive amounts of data, has demonstrated strong performance across multiple domains, including intrusion detection systems (IDS). Consequently, security researchers are increasingly integrating machine learning into IDS solutions. Numerous studies have explored this area, resulting in several approaches—most of which depend on expert-engineered features and focus on either binary classification (normal vs. attack) or multiclass classification (attack category recognition) using traditional machine learning algorithms. In this work, the publicly available UNSW-NB15 benchmark dataset to evaluate both binary and multiclass classification. To capture temporal patterns in network traffic, we propose a hybrid LSTM2LP model that combines sequential features processed by an LSTM network with static features processed by a multilayer perceptron, enabling richer feature representation. The developed hybrid LSTM2LP model accurately detects normal and abnormal classes in binary classification with an accuracy of 95%-99%, and in multiclass classification it effectively identifies Generic and Exploits attacks with an accuracy of 85%-94%.