Enhancing Cyberattack Detection Using Dimensionality Reduction With Hybrid Deep Learning on Internet of Things Environment

Abstract

Cybersecurity in the Internet of Things (IoT) ecosystem is vital to protect sensitive data, stop unauthorized access, and mitigate the risk of disruptive cyberattacks. Cyberattack recognition utilizing an intrusion detection system (IDS) is a major imperative given the increase in the number of connected devices. Advanced cybersecurity methods deploy machine learning (ML) approaches, anomaly detection, and behavioral analysis to analyze IoT network traffic for irregular patterns indicative of potential cyberattacks. The combination of feature selection (FS) and deep learning (DL) approaches in cyberattack recognition suggests a proactive and sophisticated manner to bolster cybersecurity. Leveraging DL structures like neural networks (NNs) assists the automatic extraction and analysis of intricate patterns in the difficult IoT data landscape. This paper develops an Improved Mayfly Optimization Algorithm with a Hybrid Deep Learning based Intrusion Detection (IMFOHDL-ID) approach in IoT environments. The designed IMFOHDL-ID approach's main goal is to classify intrusions and accomplish security in the IoT environment. The IMFOHDL-ID technique initially follows data normalization as a preprocessing stage. In addition, the IMFOHDL-ID technique makes use of the IMFO-based feature selection (FS) method to elect feature subsets. For IDs, the IMFOHDL-ID technique applies the Long Short Term Memory based Deep Stacked Sequence-to-Sequence Autoencoder (LSTM-DSSAE) model. Finally, the dipper-throated optimization algorithm (DTOA) was utilized for optimal hyperparameter selection of the LSTM-DSSAE method. To highlight better results of the IMFOHDL-ID model, a series of simulation analyses were performed. Extensive comparative results stated the improved outcome of the IMFOHDL-ID technique over existing approaches.