Abstract: In order to solve the problem that the accuracy of traditional Abnormal attack detection model of Internet of Things is insufficient when training samples are insufficient, decision tree (DT) is introduced to improve the deep neural network (DNN) model. The addition of decision tree makes it easier to obtain the optimal salient feature set when DNN conducts and features extraction. At the same time, features can be summarized through decision tree, and new features can be inferred through DNN's perceptual learning ability, enabling the model to identify more Internet of Things behaviors. At last, the improved model is validated by computer simulation performance, and the results show that the DT - within DNN model can attack of black holes, DDoS attacks and other Internet attack testing, and testing has relatively higher precision and recall rate and F1 score, proved that the detection model can be applied to a variety of Internet of things attack type testing, has a certain practicality.

Key words: Internet of Things, DNN neural network, abnormal attack detection, decision tree, feature screening

摘要： 针对训练样本不足时,传统物联网异常攻击检测模型精准度不足的问题,引入决策树(DT)来对深度神经网络(DNN)模型进行改进。决策树的加入,让DNN进行及特征提取时,更容易获得最优的显著特征集;同时,通过决策树还能够对特征特点进行总结,通过DNN的感知学习能力,推断出新的特征,使模型能够识别出更多的物联网行为。最后,通过仿真验证改进后的模型性能,结果表明,DT-DNN模型能够对黑洞攻击、DDoS攻击等多种物联网攻击行为进行检测,且检测具有相对较高的精确度、召回率和F1分数,证明该检测模型可应用于多种物联网攻击类型检测,具有一定的实用性。

关键词: 物联网, DNN神经网络, 异常攻击检测, 决策树, 特征筛选