Research On Intrusion Detection Methods Based On Feature Selection

Intrusion detection is one of the hot topics in network security technology.By analysing the data extracted from the network environment,it detects whether an intrusion has occurred.As computer networks continue to grow,the ever-increasing amount of network traffic poses a challenge to intrusion detection.This has resulted in intrusion detection requiring an increasing amount of data to be processed,while the processing power of the hardware has not progressed at the same rate.To improve the detection performance of intrusion detection methods,dimensionality reduction of data is an important research component.Feature selection is one of the important measures of data dimensionality reduction,which can speed up intrusion detection and improve the performance of intrusion detection.Feature selection reduces the amount of data that needs to be processed for intrusion detection by eliminating redundant features from the data.The reduction in the amount of data leads to an increase in the speed of detection and the reduction in redundant features leads to an increase in the performance of detection.A number of feature selection algorithms have been proposed for intrusion detection,but there are still some shortcomings.For example,the wrapper-based feature selection is slow and the detection performance of the subset of features found is not good enough;the filter-based feature selection can be faster,but the detection performance is low.To address the above problems,two feature selection models are proposed in this paper,and the main research work of this paper is as follows.(1)An intrusion detection method based on bionic feature selection is proposed.The method uses one-hot coding to pre-process the classification features,and performs feature selection directly in the feature space that has been pre-processed,which improves the purity of the features.At the same time,the method proposes a bionic feature selection model for extracting outstanding features to form a high-quality feature subset.The model first randomly initialises a certain amount of feature subset populations,evaluates the fitness value and then selects a subset of outstanding parental features to form a parental group.The parental group is then referred to generate a subset of offspring features and evaluate their fitness values.Finally,the subset of offspring features is sorted together with the subset of parental features according to fitness values to update the population,and so on,iteratively,to finally output the optimal feature subset.The proposed intrusion detection method is evaluated using two very popular datasets in the field of intrusion detection and compared with similar intrusion detection methods,and the experimental results show that the proposed intrusion detection method has better detection performance.(2)An intrusion detection method based on voting feature selection is proposed.The method first processes the classification features in the dataset into numerical features.The method then proposes a voting feature selection model for quickly extracting important features to form an excellent feature subset.The model first initialises a certain number of feature subsets and evaluates their fitness values,retaining some of the good feature subsets based on their fitness values.The retained feature subsets are then asked to vote on the features,and the features are ranked according to the number of votes they receive.Finally,a subset of features is generated and evaluated based on the ranking of the features,resulting in the optimal feature subset.The proposed intrusion detection method is evaluated using two datasets and compared with similar methods,and the experimental results show that the proposed intrusion detection method has better detection performance.