|
| Previous work | Method | Limitation | Solution suggestion |
|
| Suspicious fraud detection based on the white box [2] | Use the set rules to judge the behavior data of people | Rules are fixed, inflexible, and cannot accommodate new forms of fraud | It is suggested to mind the fixed process factors of fraudulent behavior and determine the fraudulent behavior by exploring the interaction between factors. |
|
| Security system based on neural network classifier [5] | Using the neural network method to classify the characteristics of mobile phone operation behavior to determine whether there is dangerous behaviour | Relying on existing operational features, it is easy to be attacked by adversarial samples | It is suggested to use the Bayesian network method to realize the detection of security by exploring the interaction between the operational feature factors. |
|
| Recognition method based on GA-SVM [8] | Divide the call record data set into subsets using the k-mean method, and then use the subsets on the model as a training set and a cross-validation set | Single data, no process factors are taken into account | It is recommended to expand the variety of data types in the dataset |
|
| Fake account identification based on the Markov network [9] | By considering the group characteristics of telecommunication fraud, the Markov network is used to identify the transaction accounts, and the fraudulent accounts are excavated. | The scope of application is small, and the whole process of fraud prevention cannot be achieved | It is recommended to integrate this method into the whole process fraud prevention system |
|
