Research On Anomaly Detection System Based On Fuzzy Clustering And Feature Selection

The rapid development of information network technology brings great security threats.How to accurately and efficiently discover anomaly behaviors in the internet has become an urgent problem.Clustering algorithm can directly establish detection models on unlabeled data sets that it is often combined with anomaly detection technology.Feature selection is generally used as a pre-processing step for classification,avoiding dimensional disasters by eliminating features of redundant interference.Anomaly data often has high dimensional and complex features,which makes feature selection widely used in anomaly detection field.For the problems existing in anomaly detection at this stage,the following three improvements have been made:1.Based on the fuzzy C-means clustering(FCM)algorithm,the innovative adaptive bat algorithm is used to optimize the fuzzy C-means algorithm.The distribution entropy and the average bit distance are added to the algorithm to adjust the traditional fuzzy C-means clustering algorithm.This algorithm effectively overcomes the sensitivity and easy to fall into local optimum,has a good anomaly detection effect.2.Based on the traditional ReliefF feature selection algorithm,the weighted KNN algorithm improved by fuzzy entropy is used to guide the feature selection,and the influence of different features in the data set on the classification is fully considered.The algorithm can select a representative subset of features,which effectively reduces the complexity of network traffic data.3.Based on the above two innovations,a new unsupervised anomaly detection system model based on fuzzy clustering and feature selection is constructed.The clustering algorithm optimized by adaptive bat algorithm is used to cluster the original data and the fuzzy entropy weighted ReliefF algorithm is used for feature selection.An abnormality detection is performed using an extreme learning machine(ELM)as a classifier.The system model can effectively overcome the dependence of the traditional anomaly detection method on the marked data set,and avoid the influence of redundant data on the detection result.Experiments on KDDCup99 and CICIDS2017 data sets show the clustering algorithm,feature selection algorithm and anomaly detection model proposed in this paper can take into account the detection rate and time complexity,have certain practical application significance.