Research On Intrusion Detection Model Based On AR Optimization

Network intrusion detection,as a security mechanism to prevent,monitor and resist system intrusion,is characterized by timeliness,dynamics and intelligence compared with other defensive security means.Research on feature selection or reduction has always been the focus of research in the field of intrusion detection,because some irrelevant or redundant features with detected objects may contain noisy data,which will affect the efficiency and accuracy of prediction.By analyzing and comparing several feature selection algorithms in detail,this paper demonstrates that Attribute Ratio(AR)performs better in algorithm implementation and optimization effect.Three intrusion detection models using AR optimization are designed,which are respectively the model combining K-Means clustering with random forest,the model combining Gaussian Mixture clustering with random forest and the random forest model classifying according to attack types.The organization structure of the three models and the principle and implementation of the main algorithms involved are described in detail.The structure and statistical characteristics of NSL-KDD data set were studied in depth,and data preprocessing and standardization were carried out according to its characteristics.The simulation environment based on Python,Scikit-learn and PySpark was built,and the simulation experiments of three models were completed.The experimental results show that the three AR optimized intrusion detection models have good performance in each evaluation index,and the performance difference between them is very small.For the detection of cross-validation set,the model combining Gaussian Mixture and random forest has the best overall prediction performance.For the detection of test data,the model that combines k-means with random forest has the best prediction performance.