Research On Harassment Call Identification Based On Neural Network

With the development of the era of big data on the Internet,telephone harassment such as advertising,insurance,credit card sales,online classes,and home purchases has emerged in an endless stream,making it difficult to prevent.And fraudulent phone calls have a significant impact on property and security.How to find out the relevant laws and effectively make early warnings against harassment and fraudulent phone calls is extremely urgent.The existing prevention and control measures are mainly divided into two ways.The first is the manual marking method on the Internet platform,which manually marks the number tag and then uploads it to the background server for data statistics.Despite the continuous development of human tagging platforms,there are still problems with insufficient number accuracy due to the subjectivity and accuracy of human tagging,and it is not possible to predict the incoming behavior of new numbers.The second way is for operators to identify and stop calls.Through call signaling and bill analysis,potential patterns of harassment and fraudulent calls can be identified,thereby achieving operations such as interception and blocking.Although the current prevention and control of harassment and fraud by operators has achieved significant results,due to relying on operators’ closed databases,there are also certain problems in accuracy and timeliness.In this context,a hybrid algorithm model based on AFSA＿PSO＿BP is proposed to predict harassing and fraudulent calls.The main work of this paper is divided into the following aspects.First,feature extraction.Through the communication system and mobile phone real-time communication interaction,protocol,Android communication architecture,harassment fraud behavior and other aspects of the analysis,summed up the relevant data information,and designed the standard data interface.Second,model and parameter selection.By analyzing the advantages and disadvantages of group optimization algorithm and BP neural network,a hybrid algorithm based on BP neural network with AFSA＿PSO is proposed.In other words,in the back propagation of BP algorithm,the artificial fish swarm algorithm is firstly used to get a satisfactory solution domain,and then the particle swarm optimization algorithm is used to quickly find the optimal weight,so as to realize the weight update.Third,design an early warning and interception system for harassing and fraudulent calls.Through the feasibility analysis to formulate the system objectives,design from the top to the bottom of the system architecture,and to be implemented,this paper in the form of a flow chart to present the realization of the logic of each function module.On the human-computer interaction interface,when the user turns on the corresponding switch,the prediction result will be presented to the user by means of mobile terminal warning.After the completion of the system design,the pressure and routine tests were carried out.The final test results met the expectations and the system design objectives were achieved.The core of this paper is the mixed neural model.This paper designs a comparative experiment of different groups.Experimental results show that the hybrid algorithm model based on AFSA＿PSO＿BP performs well in algorithm performance and prediction accuracy.It is mainly supported by experiments from the following dimensions.First,algorithm performance is compared with the performance of AFSA＿BP,PSO＿BP and AFSA＿PSO＿BP in terms of convergence and iteration times.Experimental results show that AFSA＿PSO＿BP hybrid algorithm is superior to the other two algorithms.Secondly,the error prediction performance of standard BP neural network and hybrid optimization BP neural network is compared.The experimental results show that the hybrid optimization BP neural network has better prediction accuracy.