Research Of Intrusion Detection Model Based On And Ensemble Learning Methods Assisted By Feature Selection

With the more complicated network security environment becomes,the more prominent disadvantages of the traditional intrusion detection system have been.There have been many researches about using machine learning methods for Intrusion Detection Systems.However,when study theses paper,we found that not all models have good performance both on big sample classes and small sample classes.A single basic classifier like bayes or decision tree might not have a high an accuracy for DOS attacks or normal samples.On the other hand,classifier models like Neural Networks and deep learning methods improve accuracies of DOS and normal samples,but they have poor performances for U2 R and R2 L attacks.So,how can we improve accuracies of U2 R and R2 L,at the same time,keep high accuracies for classes with large scale samples like DOS ?In this paper,we exactly adopt ensemble learning methods for building intrusion detection model,and the model is assisted by feature selection for improving performance.In order to meet the demand of high accuracies of big sample attack classes and improving accuracies of small sample attack classes,we choose stacking algorithm as the whole ensemble learning framework.Firstly,we employ random committee algorithm for combining our random tree classifiers as for the first ensemble.Then we take the model which have been dealt by random committee for primary learner of the Stacking ensemble learning framework,and bayesnet for secondary learner.Above all,we have completed the whole ensemble process.Next we choose suitable feature selection approach to deal with the experiment dataset for feature reduction.Then we use the intrusion detection model based on ensemble learning methods for handling the dataset that has been feature reduction.We have tested and verified the performance of the model that proposed by experiments.First of all,we deal with the experiment dataset by the ensemble learning model and its individual learners.We can find from the experiments' result that ensemble learning methods can improve the accuracies of small sample attack classes with keeping high accuracies of big sample attack classes.Then,we use feature selection algorithm to deal with dataset for feature reduction.Next,the dataset have been dealt is used to training the ensemble learning model,and the experiment result shows the feature selection algorithm can help the intrusion detection model based ensemble learning methods improve the performance of its small attack classes.Though the above experiment,we can draw the conclusion that the model this paper proposed is feasible.