Intrusion Detection Method Based On Machine Learning

Intrusion detection system is an important part of information and network security.With the increasing complexity of network environment,intrusion methods are constantly changing.Traditional intrusion detection technology has been unable to meet the needs of security systems.In recent years,more and more related research has combined machine learning with intrusion detection technology to improve the detection rate of detector,and reduce the false alarm rate and prediction time of the system.In this paper,two intrusion detection models based on machine learning are proposed.(1)Aiming at the problems of high dimensionality and non-linear separability of network data,an intrusion detection model ICA-MLP based on Independent Component Analysis(ICA)and MLP(Multi-Layer Perceptron)is proposed.At the beginning of the model,ICA algorithm is used to reduce the dimension of data,and the feature matrix is mapped from high-dimensional feature space to low-dimensional space.The transformed features are independent of each other.ICA eliminates feature redundancy and reduces the number of features.Then the classification is carried out by using the neural network.Experiments show that the intrusion detection algorithm based on ICA-MLP has better feature learning ability and classification ability than some machine learning algorithms.(2)Aiming at the problem of long prediction time and poor interpretability of single neural network algorithm,the first algorithm is improved.Decision Tree classifier is added between feature dimensionality reduction and neural network,and a new detection model ICA-DT-MLP is designed.The decision tree will pre-classify the data.If the decision tree considers the data as an attack,it will intercept directly,otherwise the data will enter the neural network layer for classification.Because the decision tree is a tree structure and the detection speed is very fast,the system can greatly improve the detection speed of the system.The twotier detection model also ensures the detection rate and accuracy of the system.