NETWORK INTRUSION DETECTION IN CIVIL AVIATION BASED ON IMPROVED CONVOLUTIONAL NEURAL NETWORK

Abstract

The widespread adoption of cloud computing has introduced new security challenges, such as breaches within internal civil aviation networks and deviant behavior by users. This study seeks to tackle the issue of detecting unauthorized access or attacks, as well as analyzing deviant behavior within internal networks. To achieve this, we utilized machine learning algorithms from Weka software to analyze intrusion detection data sets. Our approach involved using the naive Bayesian algorithm to identify malicious behavior by users within civil aviation internal networks and classify both normal and abnormal behavior. The results demonstrated that the naive Bayesian algorithm effectively identifies abnormal behavior with high accuracy and efficiently analyzes user behavior within internal network data for cloud-based intrusion detection computing. The evolving characteristics of wireless network traffic attacks have posed challenges Traditional intrusion detection technology has high false positive rates, low detection effectiveness, and limited generalisation ability. To increase security and identify hostile intrusions in wireless networks, we present an enhanced convolutional neural network (ICNN)-based technique. First we characterized and preprocessed the network traffic data We used ICNN to model network intrusion traffic data. CNN abstractly represented low-level intrusion traffic data as advanced features. It extracted sample features and optimised network parameters using stochastic gradient descent to converge the model. Simulation findings indicate that our suggested strategy outperformed standard models in terms of detection accuracy, true positive rates, and false positive rate. Convolutional neural networks can effectively extract characteristics from network intrusion detection data many existing methods based on them lack depth. When neural networks are deepened, problems like vanishing gradients can occur. To address these challenges This network intrusion detection solution combines an attention mechanism with DenseNet. This approach converts pre-processed network traffic data to grayscale maps and extracts features using DenseNet. This allows for deeper network architecture and prevents gradient disappearing. Experiments on the NSLKDD and UNSW-NB15 datasets show better accuracy and F1-score metrics compared to previous shallow models demonstrating their usefulness our approach.