The Research Of Kohonen Neutal Network In Telecom Frauds Forecast

With the vigorous development of the telecommunication industry, the domestic telecom market continues to expand, various forms of frauds was found in China's telecom industry, which has not only brought huge economic losses for telecom operators but also affected the further development of the telecommunications services. Facing with the growing telecommunications fraud, telecom operators must take certain measures to protect themselves. Data mining technology is currently the most powerful data analysis tool based on data warehouse, and clustering technology is now proven to be effective for detecting fraudulent characteristics of customers, thereby it is a kind of fraud prevention methods which prevents the occurrence of fraud.Through analyzing the development of telecommunication fraud at home and abroad, introducing the technology of data mining in telecommunication, combining the characteristics of the telecommunication data, this paper proposes a kind of clustering analysis method for the fraud behavior based on Kohonen neural network clustering algorithm and improves the Kohonen algorithm appropriately. It effectively improves the algorithm performance, greatly reduces the network training time. The research of this paper shows that the Kohonen neural network algorithm is effective for the analysis of predicting the telecom fraud. Simulation results show the high efficiency of the improved algorithm. It is verified that the telecom fraud prediction has great potential of application.The main contents are:(1) Analyzing the characteristics of the domestic telecommunication data and the current problems of the domestic research of telecom fraud.(2) Making deep research for the necessity and feasibility of the Kohonen neural network algorithm in predicting the telecom fraud. (3) Making deep research for the efficiency and accuracy of Kohonen neural network algorithm and applying some appropriate improvements.(4) Pre-processing telecommunication data and applying the Matlab software to simulate and analyze the experimental results.Main innovations are:(1) Kohonen neural network algorithm is successfully used in telecommunication fraud prediction, which provides an effective and predictive model of telecommunication fraud.(2) For the data that the Kohonen neural network can handle, this paper carried out an unique normalized processing method for the original telecommunication data, expanded the attributes of the telecom data sample, which has the ability to predict the telecom frauds.(3) After the study of the main parameters of Kohonen neural network algorithm, a more effective improved method is proposed in this paper.(4) Successfully applying the improved Kohonen neural network algorithm to predict the behaviors of telecom fraud, and simulation results show that the predicting efficiency has been significantly improved.