Network data trends to become large

Network data trends to become larger, more complex and multidimensional. Traditionalmachine learning methods need to manually extract a large number of features whenprocessing high-dimensional data. It makes the feature extraction process more com-plicated, increases the amount of computation and is not conducive to the real-time andaccuracy of intrusion detection. Deep learning has a good advantage in processing com-plex data because it can automatically extract better features from the data. Duringmodel training, the convolutional neural network can effectively reduce the number ofparameters and the dimension of input data, thus making up for the disadvantages ofother deep learning methods. However, because of the complex network data types, thedistribution of network data set used for training is imbalanced, SMOTE method canrebalance the data set.This dissertation improves the intrusion detection model based on the convolutionalneural network(CNN), design and implement a network intrusion detection sys-tem(NIDS) based on 3-tier architecture. According to the experimental results, the de-tection accuracy of this model in KDD99 data is up to 99.06%. Compared with thetraditional machine learning detection model and the CNN model without SMOTEmethod, the accuracy is improved.Index terms: Convolutional Neural Network, Intrusion Detection, Deep Learning, In-