Research On Wireless Lan Intrusion Detection Based On Machine Learning

With the rapid development of mobile Internet and the portability and miniaturization of smart terminals,more and more people use wireless local area networks as the preferred network connection method.However,in the use of wireless local area networks,network attacks also continue to occur and threaten people's property and life,and the forms are more and more novel and the changes are more and more diverse.Intrusion detection systems have received significant attention due to their proactive response to attacks and more comprehensive protection.Most of the current wireless LAN intrusion detection systems perform intrusion detection by maintaining a normal or abnormal signature database to perform pattern matching on the access data.However,they are inadequate when faced with a variety of new network attacks,and the cost of maintaining signature databases is huge.At the same time,the feature dimension of intrusion detection data are high,the amount of data is large,and the problem of data imbalance between categories seriously affects the performance of intrusion detection methods developed based on traditional machine learning.In view of these problems,this article studies intrusion detection in wireless local area networks and uses machine learning and deep learning methods for intrusion detection.The main contributions are as follows:1)Aiming at the problems of intrusion detection data with high feature dimensions,many new attacks,and large data volume,this paper proposes an intrusion detection model based on a deep factorization machine.The model can reduce the dimensions of the one-hot encoded high-dimensional feature data,and then input the reduced-dimensional features to a two-way factorization machine and a deep neural network at the same time to simultaneously extract the low-order combined features and deep features of the data.The results are fused and classified using the Softmax classifier.The experimental results of this method on the KDD CUP 99 data set have been improved compared to the mainstream intrusion detection machine learning algorithms,and the accuracy rate has reached 93.4%.2)Aiming at the imbalance between the categories of intrusion detection datasets,this paper proposes an integrated learning method based on the Deep Factorization Machine(DeepFM)based on the model in the previous chapter.This paper first improves the SMOTE-Borderline algorithm and proposes an improved SMOTE-Borderline method to prevent the data center of gravity from shifting when the replication multiple is high.Then use the improved SMOTE-Borderline oversampling algorithm and Easyensemble undersampling algorithm to sample and generate the training set,and then use the Bagging integrated learning algorithm with a deep factorization machine as the base learner to train.Experiments using the KDD CUP 99 dataset prove that the algorithm can effectively solve the problem of data imbalance and improve the model classification effect.3)Design and implement a wireless LAN intrusion detection system based on the algorithm in this paper.The system is divided into six modules:data acquisition module,data processing module,intrusion detection model training and detection module,data storage module,access control module,user interaction and visualization module,and system architecture design using MVC mode.The core method used in the intrusion detection model training and detection module is the ensemble learning method based on DeepFM model proposed in this paper.Then use Python,bootstrap,Flask,Keras,Tensorflow,Scapy,Pandas and other technologies to implement the encoding of this system and carry out experiments to verify the effectiveness of the system.