Study Of SVM Application To Customer Fraud Detection In Telecommunication

With the development of the telecommunication market, customer fraud behaviors are growing rapidly. Fraud detection systems detect, prevent fraud behavior by analyze the customer behavior, use the advanced technology such as data mining and pattern recognition. The detection systems combining information and consumption behavior of customers are constructed to discover and prevent user's fraud. So the carriers could prevent fraud behavior, improve the relationship between customer and enterprise, which can solve the difficulty in the management of customer relationship.The true intention of user's can't be get directly only reflected in the call detail record. So the calling record is studied to descript and generate user's behavior model. The main purpose of our study is to use user's profiling and classification technique to classify various custom groups, and assigned the unknown customer to correspond group for fraud detection.The harmfulness of the customer's fraud and necessity of telecommunication custom fraud is first discussed. The related methods which fraud detection use introduced, such as machine learning, statistic theory and so on. Subsequently classification based on one class support vector machine is proposed to apply in framework of detection fraud. Data which sampled by random selection method is classified using one class support vector machine at the first stage, then using support vector machine at the second stage. Secondly appropriate attributes of user behavior are selected for user profiling. The process is consisted of data collection, data cleaning, data transformation and feature selection. At the end of paper Fraud detection and validation process is highlight. Experiment results shows the actual fraud detection system based on SVM in telecommunication have important market value. The system can discover potential fraud behavior, and identified a large number of small scale but large dispersion telecommunication fraud.