Abnormal User Prediction And Recognition Based On Data Mining Method

With the increasing proportion of computer networks in daily life,people slowly shift their gaze from the web to the mobile side.Cashless payment,APPand other mobile products have gradually become a necessity in life.There has been a fierce struggle between the APP products for the user volume.As the profit of APP increases,the black industry will follow.For enterprises,while creating revenues,they must also prevent black industries from using illegal means to steal profits.In order to effectively combat the black production method,we need to understand the basic characteristics of the user and the usual means of black production,and find the essential difference between the normal user and the abnormal user.Although invite-reward have less interest,and the behavior of the black industry is not obvious.However,if an abnormal user can be accurately identified,different confrontation strategies are given to users of different risk categories.It is of great significance to the company to extend the theoretical framework to other businesses and ensure business benefits.This paper is based on the data mining method to identify the abnormal users of online user operation behavior.Firstly,the extracted feature rules are analyzed by Apriori algorithm and K-means clustering.Evaluate the rationality of feature rules and rule sets.Based on confidence and clustering centers,optimize feature rules,save space and reduce the cost of high scoring due to rule stacking.Secondly,predict the risk level of online user operation behavior,This paper compares the prediction results of the SVM model,the Logistic Regression model and the LightGBM model.Finally,a better SVC algorithm based on SVM model is selected.SVC is more accurate and faster than other algorithms,and the model with the kernel function Rbf is better,with a correct rate of 0.9715.Finally,a Bayesian network based on trust relationship is proposed to identify hidden black industry users in online users.However,the effect of the model is not very good.The reason is that the user's network is not very strong.The number of people around the network of a user is less than 1500,which is not convenient for the analysis and calculation of the Bayesian network.However,this method can provide other businesses with the idea of identifying hidden users and provide decision-making basis for abnormal user interception.