Open AccessStudy on Intrusion detection model based on improved convolutional neural network
Author(s) -
Kun He
Publication year - 2021
Publication title -
journal of physics. conference series
Language(s) - English
Resource type - Journals
SCImago Journal Rank - 0.21
H-Index - 85
eISSN - 1742-6596
pISSN - 1742-6588
DOI - 10.1088/1742-6596/1865/4/042097
Subject(s) - computer science , data mining , pooling , convolutional neural network , intrusion detection system , artificial intelligence , layer (electronics) , process (computing) , feature (linguistics) , data set , constant false alarm rate , set (abstract data type) , artificial neural network , machine learning , pattern recognition (psychology) , network security , linguistics , chemistry , programming language , operating system , philosophy , organic chemistry
With the rapid development of information technology and cyberspace, information interactions between networks are becoming more frequent. At the same time, cyberattacks are posing more and more threats to network security through intrusions into computer systems or information systems. To address these problems, this paper proposes an improved convolutional neural network-based intrusion detection model (ICNN-IDS) to determine and classify specific types of intrusions after feature extraction and analysis of different network flow. The paper introduces the basic components of neural networks, including the convolutional layer, pooling layer and fully connected layer. Next, the experimental data set acquisition and pre-processing process are introduced, followed by the structural setup, the specific tuning process of the model, and the optimization of the model parameters are evaluated through experiments. In order to optimize the input feature matrix, the model adds a neuron mapping layer before the convolutional layer to convert sample data from 1D to 2D for the improvement of the model. The experimental results show that ICNN-IDS achieves 99.35% detection accuracy and a low false alarm rate of 0.21% on the KDD99 dataset with optimal parameter settings, which has significant improvement over existing detection models.