基于异常行为数据流的加权铰链分类算法研究

Research on Weighted Hinge Classification Algorithm Based on Abnormal Behavior Data Stream

为了能够快速检测、识别异常行为数据流,并对异常行为数据流的内容准确定位,通过研究决策树算法、逻辑回归算法、神经网络算法、铰链分类等深度学习算法的基础上,文章提出基于指数加权和动量的小批量梯度下降铰链分类算法。该算法在铰链分类算法的基础上增加指数权值方法和动量梯度下降方法,指数加权方法可以实现对损失函数下滑速度的加快,即对异常行为数据流快速定位和识别;动量梯度下降方法可以实现对损失函数结果值进行动态偏差校正。通过该算法可以实现对异常行为数据流内容准确定位及快速识别,解决异常行为数据流内容识别的误报率和漏报率较高的问题,保障数据的安全性和完整性。通过仿真实验,利用计划长度比(schedule length ratio,SLR)、下降速度等指标分别验证,表明该算法的性能和收敛速度等方面都略高于其他铰链分类算法。

In order to quickly detect and identify abnormal behavior data streams,and accurately locate the content of abnormal behavior data streams,Based on the research of decision tree algorithm,logistic regression algorithm,neural network algorithm,hinge classification and other deep learning algorithms,a small batch gradient descent hinge classification algorithm based on exponential weighting and momentum is proposed.This algorithm adds an exponential weight method and a momentum gradient descent method on the basis of the hinge classification algorithm.The exponential weight method can accelerate the decline of the loss function,that is quickly locating and identifying abnormal behavior data streams.The momentum gradient descent method can realize real-time deviation correction of the result value of the loss function.The algorithm can realize the accurate positioning and rapid identification of abnormal behavior data stream content,solve the problem of high false positive rate and false negative rate of abnormal behavior data stream content recognition,and ensure the security and integrity of data.Through simulation experiments,SLR,Speed and other indicators show that the performance and convergence speed of this algorithm are slightly higher than other hinge classification algorithms.