Research On Intrusion Detection Model Based On Big Data Features

With the rapid development of network technology,the Internet has penetrated into every step of people's daily production and life.As the importance of the Internet continues to increase,security issues have become increasingly acute.In the context of big data,network security issues are characterized by new models,large scale,and high concealment.Therefore,research on intrusion detection models based on big data features has received extensive attention on the field of network security,and the model can be applied to military,industrial,and communications fields.Aiming at the deficiencies in the existing intrusion detection models,such as excessive reliance on feature libraries,single model optimization methods,and inability to adapt to the existing big data environment,this thesis proposes an intrusion detection model based on big data features.It aims to construct intrusion detection models for network intrusion data sets of different types,scales and structures.This thesis comprehensively uses dynamic selection,data mining,weighted integration and other technologies to realize network intrusion detection in a big data environment.To deal with the problem of network intrusion,the main research work of this thesis includes the following parts :1.In order to improve the accuracy of the intrusion detection model and solve the problem of the traditional model overly relying on the feature library,a two-layer integrated intrusion detection model is proposed.By improving the classic Stacking integration algorithm and introducing the homogeneous integration algorithm,the selection range of the classification algorithm is expanded,and the accuracy of the intrusion detection model is improved.At the same time,this model summarizes the characteristics of the data through the learning of big data and uses this as the basis for intrusion detection.The grouping experiment method is adopted,and the integrated classification algorithm is selected to ensure the high accuracy of the model while also having better performance.Experiments show that when the number of homogeneous ensemble algorithms used in the two-layer ensemble model is greater than or equal to 4,the accuracy growth tends to be stable.The 4 algorithms with the highest accuracy are selected for detection,and the accuracy is about 0.980.2.In order to solve the problem of single optimization method of intrusion detection model and insufficient utilization of big data features,this thesis proposes a dynamic selection weighted intrusion detection model based on big data features.In order to complete this model,first use dynamic selection based on classification algorithm evaluation indicators.The algorithm changes the single selection criteria in the traditional dynamic selection algorithm.On the basis of using the confusion matrix to calculate the accuracy and evaluation indicators of the basic classification algorithm,the evaluation indicators are clustered into clusters,and each is selected according to the accuracy.The best algorithm in the cluster.Secondly,aiming at the problem of insufficient utilization of big data features,we propose a data-classification algorithm applicability index,and use a subjective and objective weight combination calculation method to calculate model weights,and use distance functions to combine subjective and objective weights to generate combined weights.And bring it into the weighted voting algorithm to calculate the intrusion detection result.Experiments show that this model can dynamically select classification algorithms based on the characteristics of the data set,while optimizing the number of classification algorithms used,to a certain extent,solves the problem of redundancy or insufficiency of classification algorithms in the integration process.The intrusion detection models that combines the characteristics of the classification algorithm with the attributes of the data set to improve the performance of the intrusion detection model.The experimental results show that the detection accuracy of the intrusion detection model constructed for the KDD Cup99 and CIC-IDS2017 datasets can reach 0.988 and 0.996,respectively,and the F1-Measure of the model can reach 0.93 and0.96,respectively.