基于多重检验特征选择的物联网设备识别

IoT Device Identification Method Based on Hypothesis Testing Feature Selection

准确识别连接到组织网络的物联网设备是保护组织网络安全的有效方法。目前基于简单特征的识别任务效果难以保证,且较少有针对设备识别任务中特征选择问题的系统性研究。针对物联网设备识别任务中的特征选择问题,提出了一种物联网设备特征选择方法。该方法利用Kruskal-Wallis检验先选出总体分布存在差异的特征,然后使用组间差异检验方法量化特征出现频率和组间差异,最后利用特征重要性从多维特征中筛选出存在组间偏差的最优特征子集。在公开数据集上进行实验,比较该最优特征子集在常见机器学习方法与特征选择方法中的实验结果,验证了物联网设备特征选择方法的可行性与有效性。

Accurate identification of Internet of Things(IoT)devices connected to the organization network is an effective way to maintain the organization network security.The effect of recognition tasks relying on simple features cannot be guaranteed,and systematic research on feature selection in device identification tasks is rare.To address the feature selection problem in the identification task of IoT devices,a feature selection method for IoT devices is proposed.This method uses the Kruskal-Wallis test to first select the features with differences in the overall distribution.Then the feature existence deviation extent and the inter-group deviation are quantified using the inter-group difference test method.The optimal feature subsets with inter-group deviations are screened out from multi-dimensional features by feature importance.Experiments are finally carried out on public datasets to compare the experimental results of the optimal feature subset in common machine learning methods and feature selection methods.The experimental results verify the feasibility and effectiveness of the feature selection method for IoT devices.