Research On Anomaly Detection Model Of IOT Network

In today's Internet of Things era,massive amounts of data are transmitted on the Internet of Things network.In the Internet of Things network,anomalies can be ubiquitous,such as DoS attacks,probe attacks,unauthorized access and so on.These anomalies have brought big problems to the entire Internet of Things environment.At the same time,with the swift growth in the amount of data transmitted over the Internet of Things network,the network data collected is often in a state of imbalance.If anomaly detection is done on the IoT network data at this time,due to the large amount of data and the imbalance of the data set,our anomaly detection accuracy rate may be affected and the detection system may not have time to respond,or will even fail.To solve the above problems,academic circles have focused their research on feature selection methods and auto encoder methods.The former can perform feature selection for the Internet of Things network big data to reduce dimensionality,reduce computational complexity,select features that lead to anomaly detection,and improve the efficiency of anomaly detection;the second can learn deep features in high resolution.Dimensional data sets,combined with the cost sensitivity method,processing unbalanced data sets,can effectively improve the accuracy of our IoT network anomaly detection.In general,the main contributions and specific research points of this thesis include the following aspects:(1)In view of the complexity of the characteristics of the Internet of Things network data set,the massiveness of the data,etc.,direct training on this data means high time overhead and low efficiency of anomaly detection,and high-dimensional features of the dataset are reduced Dimension is very necessary Therefore,we study an anomaly detection model based on the combination of selection and classification of characteristics:F?MIC and random forest anomaly detection model.We verified that the proposed anomaly detection model can effectively improve the accuracy,precision,recovery rate and F1-score of the anomaly detection of IoT network attacks through real data sets.(2)There is an imbalance problem in the data set of attacks on the Internet of Things network.Therefore,the research is based on the cost-sensitive deep autocoder CDAE anomaly detection model.We verified the accuracy,precision,recovery and F1-score of this model for some attack classes and most attack classes using real data sets.(3)Design and implement a subsystem to implement anomaly detection in the Internet of Things network.Based on the results of the above research,the proposed F?MIC and Random Forest Anomaly Detection Model and the CDAE Anomaly Detection Model are applied to the Sichuan Changhong project subsystem to visually display the detection results of anomalies.This thesis further enhances the detection capabilities of the IoT network anomaly detection model and solves some of the existing problems of existing IoT anomaly detection technology,which has high application value and practical importance.