A Novel Multi-Wavelet Oriented Auto-Encoder for Intrusion Detection in IoT System

Abstract

IoT devices become more integrated into daily life, they are increasingly vulnerable to cyberattacks, compromising user confidentiality. Although existing intrusion detection techniques for IoT systems have been developed, they often fail to accurately classify attacks. This paper presents a novel approach for detecting intrusions in IoT devices by combining advanced feature extraction and deep learning techniques. The proposed method first pre-processes dataset images to enhance data quality by filtering out irrelevant information. A unique Aquila Optimized Convolutional Neural Network (AO-CNN) is then applied to extract optimal features. The proposed AO-CNN incorporates an optimization technique called Aquila Optimizer that fine-tunes the CNN's ability to extract more relevant and discriminative features from the IoT data. For attack detection, an innovative Attention-Based Multi-Wavelet-Oriented Autoencoder (AMV-AE) is designed for more precise attack classification. The Attention Mechanism is the model to focuses on the most relevant features, ensuring that the key patterns indicative of an attack are not lost during the detection process. Multi-Wavelet Transform enhances feature representation by capturing both time and frequency domain characteristics of the data, making it particularly effective in identifying subtle anomalies that may indicate an intrusion. The key novelty of this approach lies in the integration of AO-CNN for feature optimization and AMV-AE for superior detection accuracy. Evaluated on the NSL-KDD dataset, the model achieves a recall of 98.49% and an accuracy of 99.35% while demonstrating reduced inference time and memory usage, outperforming existing methods.