Research On Iot Device Intrusion Detection Algorithm Based On Machine Learning

With the wide application of the Internet of things,the security of Io T devices has attracted great attention.Io T devices are vulnerable to a variety of attacks.In order to avoid Data breach,malicious intrusion,denial of service availability and other malicious attacks,effective defense strategies need to be deployed to ensure that they are in a safe operating environment.Anomaly detection technology can detect unauthorized activities on Io T devices,ensuring the security of Io T networks and devices.Machine learning technology can use key features to construct traffic data detection models,and it has broad application prospects in anomaly detection of Io T devices.Therefore,this thesis combines machine learning technology with anomaly detection system,and proposes an anomaly detection algorithm for Io T devices based on Federated learning and an adversarial defense algorithm based on Ensemble learning,which can effectively deal with abnormal behavior of devices.The specific research points of this thesis are as follows:(1)Aiming at the problems of high training cost and low accuracy of the global model caused by inconsistent performance of local models in intrusion detection algorithms using Federated learning technology,this thesis proposes a lightweight device anomaly detection algorithm based on Federated learning.This algorithm uses Variational autoencoder and Light GBM at network edge nodes to reduce the dimensions of data and extract features,which eliminates redundant features and reduces model training time;When uploading model parameters,a dynamic weighted gradient update algorithm was used to reduce the impact of poor local model performance on the global model during the training process.The experimental results show that compared with the control group,the precision,recall,accuracy,and F1 score indicators of the proposed algorithm in this thesis have significantly improved.(2)Machine learning algorithms are susceptible to adversarial attacks,which generate adversarial samples that greatly reduce the accuracy of the model.To solve this problem,this thesis proposes an adversary training defense algorithm based on Ensemble learning.This algorithm uses six typical adversary attack algorithms to generate confrontation samples,adds confrontation samples to the training set,and improves the accuracy of the model;By integrating multiple detection algorithms to construct the final anomaly detection model,the AUC value of the detection algorithm has been improved.The experimental results show that under adversarial attack interference,the accuracy and AUC value of the detection model are improved.