基于决策树和混合神经网络的大数据攻击增量检测研究  

INCREMENTAL DETECTION FOR BIG DATA ATTACKS BASED ON DECISION TREE AND HYBRID NEURAL NETWORK

作　　者：谭继安[1] Tan Ji’an(Dongguan Polytechnic,Dongguan 523808,Guangdong,China)

机构地区：[1]东莞职业技术学院,广东东莞523808

出　　处：《计算机应用与软件》2022年第7期329-335,349,共8页Computer Applications and Software

基　　金：广东省高等职业教育教学质量与教学改革工程项目(GDJG2019006);东莞职业技术学院国家双高计划电子信息工程技术专业群专项经费资助(ZXF020)。

摘　　要：大数据攻击检测是一种不平衡数据的分类问题,传统的深度学习算法对此类问题容易发生过拟合,且计算时间较长。对此,提出基于决策树和混合神经网络的大数据攻击增量检测模型。模型通过卷积神经网络提取数据的特征,基于长短期记忆网络学习所提取特征之间的依赖关系,防止出现梯度消失问题。设计了基于决策树的神经网络增量学习算法,能够识别出数据的细粒度类标签。实验结果表明,混合神经网络有效地缓解了过拟合问题,提高了模型的计算效率,同时也验证了增量学习的有效性。The attack detection for big data belongs to the classification problem of imbalanced data. The traditional deep learning algorithm is prone to over fitting for such problems, and it costs long computational time. In view of this, we propose an incremental detection model for big data attacks based on decision tree and hybrid neural network. The model extracted data features through convolutional neural network. It learned the dependency relationships between input features based on long short-term memory network, and avoided gradient disappearance. We designed an incremental learning algorithm for neural network based on decision tree, which could recognize the fine-grained class label of data. Experimental results show that the proposed hybrid neural network effectively reduces the over-fitting issue, and improves the computational efficiency. It also verifies the effectiveness of incremental learning.

关 键 词：深度神经网络 大数据 数据安全 卷积神经网络 决策树 增量学习 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]