A Fraudulent Transactions Simulation Method Based on Genetic Algorithm

Abstract

When using machine learning or other methods to construct the fraud detection models, the banking industry faces such problems: the number of fraud transactions data is too small, which affect the training of anti-fraud model and the detection effect of fraud transaction. This paper proposed a data simulation algorithm based on genetic algorithm (GA-DS). By studying the feature of real fraudulent transactions, we designed the crossover mutation and other genetic operators, explored the suitable fitness function that can evaluate the quality of simulated data, and generated simulated data satisfying the characteristics of the original transaction. The experiment result shows that mixing the simulated data and the original data into the training can improve the detection ability of anti-fraud model.