Development of PCCNN-Based Network Intrusion Detection System for EDGE Computing

Abstract

Intrusion Detection System (IDS) plays a crucial role in detecting and identifying the DoS and DDoS type of attacks on IoT devices. However, anomaly-based techniques do not provide acceptable accuracy for efficacious intrusion detection. Also, we found many difficulty levels when applying IDS to IoT devices for identifying attempted attacks. Given this background, we designed a solution to detect intrusions using the Convolutional Neural Network (CNN) for Enhanced Data rates for GSM Evolution (EDGE) Computing. We created two separate categories to handle the attack and non-attack events in the system. The findings of this study indicate that this approach was significantly effective. We attempted both multiclass and binary classification. In the case of binary, we clustered all malicious traffic data in a single class. Also, we developed 13 layers of Sequential 1-D CNN for IDS detection and assessed them on the public dataset NSL-KDD. Principal Component Analysis (PCA) was implemented to decrease the size of the feature vector based on feature extraction and engineering. The approach proposed in the current investigation obtained accuracies of 99.34% and 99.13% for binary and multiclass classification, respectively, for the NSL-KDD dataset. The experimental outcomes showed that the proposed Principal Component-based Convolution Neural Network (PCCNN) approach achieved greater precision based on deep learning and has potential use in modern intrusion detection for IoT systems.

Functional Pattern-Related Anomaly Detection Approach Collaborating Binary Segmentation with Finite State Machine

“…To evaluate the proposed approach, we perform some compared experiments by using the typical SWaT dataset, and the experimental results fully verify that: (1) our approach can achieve a relatively ideal detection precision and recall rate, both of which exceed 90% for the majority of attack types; and (2) the high F1-score also explains that our approach has fine detection stability. Although our approach exhibits the relatively excellent detection ability to identify the process control-oriented threat, some problems still deserves further investigation, mainly including: firstly, for some complicated control and monitoring process that involves lots of critical control devices, the pattern association analysis may become one bottleneck problem due to its degraded performance; secondly, there is still room to improve the detection precision for some special attacks; thirdly, because one complicated control and monitoring process may generate one large-scale functional state transition model, the overfitting and model tuning is further required to guarantee the fine enough detection performance [26,44,45].…”

“…Moreover, AI algorithms have excellent self-learning and self-adaptation abilities, which are particularly suitable for analyzing the implicit data correlation and multi-dimensional behavior changes in industrial automation control [19]. More specifically, the AI algorithms applied in ML-based anomaly detection largely consist of the traditional machine learning algorithms and the deep learning algorithms: (1) the traditional machine learning algorithms, such as Neural Network [20], Support Vector Method [21], Clustering Algorithm [22], Genetic Algorithm [23] and Decision Tree [24], can offer a fast and convenient detection service due to their lower computational complexity, and are well-suited for the small sample data handling; (2) the deep learning algorithms, such as Long Short-Term Memory [25], Convolutional Neural Network [26] and Recurrent Neural Network [27], can possess excellent learning and adaptive abilities to exploit some potential correlations in a large quantity of data, and jointly execute and optimize the feature representation and classification training by designing some optimization algorithms [28,29]. However, these algorithms frequently require more computing resources and greater computing capability due to their high complexity, and may have an adverse effect on detection efficiency.…”

Swin-Transformer -YOLOv5 for lightweight hot-rolled steel strips surface defect detection algorithm

“…Moreover, the Faster R-CNN de-signed by the team further improves the real-time performance of the detection system [ 17 ]. Haq et al proposed Principal Component-based Convolution Neural Network (PCCNN) approach using CNN to detect intrusions, achieving greater precision based on deep learning [ 18 ]. YOLO enables CNN-based detection methods to be applied to real-time industrial scenarios by treating the bi-classification task as a regression problem.…”

ICIF: Image fusion via information clustering and image features

“…Besides, the title called deep-learning methods include both Machine learning (ML) and Deep learning (DL) models that have been broadly implemented in many areas such as image fusion [27,28], agricultural surveillance [29,30], environmental monitoring [31][32][33][34][35][36], sentiment analyses [37], medical image processing [38], and cyber security [39][40][41]. Moreover, the implementation of the deep learning methods can be divided into (1) training models, and (2) non-training models.…”