A Spam Detection Model Combined Negative Selection With Support Vector Machine

Spam has gradually become the main media of criminals to attack fraud users.The existing spam has various types of content and camouflage means.In order to protect the normal use of e-mail users,spam detection is a task which brooks without delay.Spam detection technology is divided into two categories:First,the black and white list as the representative of the knowledge-based approach,they are based on the list of illegal sources to identify spam.But there is a high omission rate for the unknown source of spams.Second,support vector machine and some other algorithms as the representative of the machine tearning methods,they build a model through the training of sample according to the different features of normal mail and spam.The classification of the model depends entirely on the training set,so the way of manually obtaining the training set has brought greater uncertainty for the detection.And the model has weak adaptivity.Inspired by the computer immune system,according to the negative selection algorithm in the immune system-it has the advantage of non-self-discrimination based on limited self and strong adaptivity,this paper proposes a method combined support vector machine and negative selection algorithm to build a dynamic feedback adaptive spam model Through the negative selection algorithm it will detect the miss new spam and dynamically update training set to re-generate the detector,therefore solve the problem of uncertainty by artificial selection,so that spam training and testing become a dynamic balance process.The specific contents of this paper are as follows:1)Describe the current situation of spam It describes the harm and development of spam,and the dynamic variability of existing spam in the forms of content and camouflage.By summarizing the existing spam detection technology and solution,there is no self-adaptability in the process of spam detection,and there is a problem of manual disturbance in the selection of training set.2)Proposed a model combined negative selection algorithm with support vector machine for spam detection.The model includes four parts:mail preprocessing,detector generation,model detection and the feedback updating of SVM classifier.It can not only use the mature detectors generated by negative selection to adaptively detect unknown new spams,but also use new spams to update training set dynamically,and generate a new SVM classifier without human intervention3)With the help of the public data set Enron Email Dataset to build a spam detection experiment system,this paper desig:ned two groups of experiments to verify the design of the SVM-NSA spam detection model of the actual classification effect,and with a single algorithm model for horizontal comparisoa Experiment results show that the model can detect more spams than the single algorithm model without affecting the recognition rate of normal mail.