Research On Adaptive Security Vulnerability Detection Method Based On Machine Learning In IoT Environment

With the rapid development of Io T technology and smart devices,the security of Io T devices and data has become more prominent.Security vulnerabilities in Io T devices may lead to serious cyber security incidents or privacy data leakage.To address these issues,in recent years,researchers have developed various machine learning and deep learning models for the field of Io T device security vulnerability detection.In practical applications,the detection effectiveness of a single model in complex network environments and high data traffic is often limited by the actual scenario.Single machine learning models also suffer from poor scalability,bias,and inability to detect vulnerabilities that do not exist in the training set.In addition,how to choose the right model to cope with different application scenarios is also a pressing security challenge that needs to be addressed today.In response to the above problems,this thesis researches machine learning-based vulnerability detection technology in the Io T environment and proposes an adaptive security vulnerability detection model based on machine learning,with the main research content as follows:In terms of model decision-making,this thesis proposes a performance decision model based on decision trees.The model can comprehensively select the detection model that should be retrained and deployed under the current environment based on the properties preferred by the user(such as time,efficiency)and the properties of the candidate model(such as accuracy,precision).First,this thesis introduces decision metrics such as accuracy,precision,recall,F1 score,time and space,memory consumption,and evaluates the model’s performance from different perspectives.Then,based on the user’s preference for these metrics,weights and tendencies are assigned.Finally,based on the decision tree,different properties are calculated for decision-making and weight calculation to output the optimal result.In terms of model adaptive training,this thesis adopts an automated machine learning algorithm to optimize the hyperparameters and structure of the model through a search space,and continuously updates the deployed environment’s vulnerability data in the database,so that the model can be quickly retrained with new data when new vulnerabilities arrive.Firstly,based on the database design of the data processing unit for capturing and storing abnormal traffic,abnormal data is obtained for subsequent model training.Secondly,this thesis uses the (8(80)4)0)(67)4)9)2)algorithm to design and implement the hyperparameter search algorithm,defining a hyperparameter search space for different types of machine learning models,thus adaptively finding the model structure that best suits the dataset during the machine learning model training phase,achieving the goal of optimizing the model training effect.In conclusion,this study conducted a comparative experiment based on automatic machine learning and manual adjustment,using the 23 dataset collected from real Io T devices vulnerable to attacks to train models.The experiment evaluated the results and intermediate decision indicators of the two approaches.Through the analysis based on the decision tree unit,the experiment found that the model training based on automatic machine learning achieved better detection results,and the decision tree evaluation algorithm could also select the appropriate model output according to the user’s preferences.Therefore,it can be proven that automatic machine learning algorithms have better applicability and effectiveness in vulnerability detection in Io T environments,and can provide better solutions for Io T security issues.