Intrusion Detection Model Of Internet Of Things Based On Light Gradient Boosting Machine

The Internet of Things（Io T）realizes interconnection between terminal devices,sensors,and other devices according to certain protocols through the Internet or telecommunication network.In recent years,the scale of the Io T has expanded faster and faster.From smart homes to smart cities,the role of the Internet of Things is becoming more and more important and will become an indispensable part in the future.However,as more and more sensitive information is uploaded to Internet of things devices,Internet of things devices have become high-value targets for cyberattacks.The intrusion detection system（IDS）detects the existence of attack behavior by detecting the traffic data transmission in the network in real time,and sends an alarm to the network security personnel to achieve the purpose of protecting the security of network equipment.Various types of IDS based on machine learning or deep learning models have been used successfully.Therefore,more and more Io T security researchers pay attention to how to use IDS to protect the Io T devices.Compared with traditional network devices,most Io T devices have limited computing power,storage,and other resources because they need to be deployed in harsh environments without human maintenance.This makes the Io T devices vulnerable to attack and difficult to get effective protection.However,traditional IDS often focus on the improvement of detection capabilities,so that they often have too high time and space complexity and are difficult to deploy in Io T devices.To solve the above problems,this paper proposes an Io T intrusion detection model based on the light gradient boosting machine（Light GBM）.First,a feature extraction model is constructed using a one-dimensional convolutional neural network（CNN）.High-dimensional network traffic data often has redundant information,which increases the burden of intrusion detection systems.CNN learns effective information from high-dimensional data through convolutional layers,reduces feature dimensions and enhances the model’s anti-overfitting capability through pooling layers.Therefore,the feature extraction model constructed by one-dimensional CNN can extract effective information from high-dimensional network traffic data and reduce the feature dimension of network traffic data,thus reducing the amount of data that the subsequent classification model needs to process.Then,the Light GBM is used to build a classification model to classify the network traffic data after feature extraction,to detect the type of network traffic data.By speeding up the construction of decision trees and reducing the number of samples and features that need to be processed,the Light GBM reduces memory consumption while inheriting the advantages of high accuracy of gradient boosting trees,making it more lightweight.Using it as a classification model can greatly speed up the operation speed without reducing the detection ability of intrusion detection system.Finally,simulation experiments are carried out on the NSL-KDD,NSL-KDD^k and UNSW-NB15 datasets for the proposed model and comparation models.The results on training time and prediction time show that the proposed model has a higher degree of lightweight.The results on the evaluation metrics such as accuracy show that the proposed model has stronger detection capability.