Intrusion Detection Based On Deep Learning

With the rapid increase of the current network bandwidth and complexity of the network topology, network intrusions become more and more diversified. The consequent massive data traffic and diversification intrusion alarm data features has become an important factor plaguing the intrusion detection system performance. Faced with such a large data flow and feature information, how to choose effective feature as a standard to judge the invasion is a major challenge in the field of intrusion detection.In the intrusion detection based on feature processing in the past, feature processing is often characterized by a simple feature selection or extraction, which can’t enhance the performance of intrusion detection system obviously. In recent years, the proposed of deep learning technology and its successful application in many areas makes people pay more and more attention to its excellent ability of features learning. In order to improve the plight of intrusion detection and make full use of the excellent features learning ability, a new method named intrusion detection based deep learning is proposed in this paper.Through comprehensively analysis of deep learning and intrusion detection, this paper proposes a hybrid intrusion detection model based on depth structure. First, the model uses unsupervised features learning with a multi-layer structure depth, the invasion data with high-dimensional and non-linear mapping to the low-dimensional space, and construct a mapping relationship between high-dimensional and low-dimensional, then change the model using the fine-tune algorithm to achieve the best features of expression, finally identify the invasion data by using classification methods. Then, do experiment which uses NSL-KDD dataset to compare the ability of features learning under deep structure model in this paper. Experimental results show that the model of depth structure is more efficient than shallow structure in the the ability of feature learning, such as the DBN which has the122-100-80-50-25-5structure increases4.53%than the DBN which has the122-50-5structure in classification rate. Results also shows that while the structure at the same depth, DBN has more better characteristics skills, such as compared with Stacked Auto-encoders, it improved2.56%, compared with Convolutional Neural Networks, it improved2.39%.On this basis, this paper proposes a hybrid intrusion detection model based on the deep belief networks. In this model, DBN was used to process the feature which has five-layer structure, next using support vector machine to identify the intrusion data. Meanwhile, this paper do a comparison experiment between this model and traditional method. Results shows that, in the area of classification, the model compared with SVM and bayesian network have greatly improved the processing time, and recognition rate is higher; in the area of feature learning, DBN is more better than traditional method, such as Gain Ratio and Principal Component analysis. In general, the feature learning ability of depth structure is more efficient and higher accuracy than under the traditional method, especially in such a feature-rich dataset.