Research On Intrusion Detection Methods With Incomplete Data Sets

In recent years,deep learning has been widely used in computer vision and natural language processing,and has achieved remarkable results.A great advantage of deep learning is that it can automatically learn and solve complex nonlinear problems.Combining deep learning with intrusion detection is one of the research directions at home and abroad.However,in order to ensure the generalization ability of deep learning model,the training process needs a lot of data,which leads to two problems in deep learning algorithm:one is how to accurately detect minorities when the distribution of data sets is uneven,the other is how to detect unknown attacks without samples.In order to solve these two problems,the research on intrusion detection methods with incomplete data sets is conducted.The main work and innovation are as follows:1)An intrusion detection method for minorities based on combined sampling and CNN(convolutional neural network)ensemble is proposed.The CNN based intrusion detection model can alleviate the problems that traditional intrusion detections faced.At the data level,a combined sampling method is proposed to adjust the data balance between different classes and reconstruct data sets;at the algorithm level,the base model is trained and fused many times based on the idea of ensemble learning.In this way,the problem of poor detection performance of minorities can be solved.2)An intrusion detection method for classes with no sample based on sparse autoencoder is proposed.The natural language descriptions of all attacks are transformed into semantic vectors,and the mapping between the original features and the semantic vectors is obtained by sparse autoencoder.The semantic vectors of unknown attacks with no sample are obtained by the mapping,and the similarity calculation is carried out to find the closest vector,so as to complete the detection of unknown attacks.3)The verification experiments of intrusion detection methods with incomplete data set are conducted.Experiments are conducted on a public data set to test the detection performance of the intrusion detection method for minorities based on combined sampling and CNN ensemble,and compared with other four algorithms to prove its effectiveness.Experiments are designed to test the intrusion detection method for unknown attacks based on sparse autoencoder,to verify whether the proposed model can detect unknown attacks without sample,and to verify the feasibility of this model.The intrusion detection method for minorities based on combined sampling and CNN ensemble can ensure the overall accuracy and improve the detection performance for classes with few samples significantly,which proves its effectiveness.The intrusion detection method for unknown attacks based on sparse autoencoder has the ability to detect unknown attacks,which proves its feasibility.