Hybrid Genetic Algorithms and Neural Networks for Credit cards fraud detection

Fraud detection is one of the applications of data mining techniques which is used in banks and credit institutions. Preparing a proper model to detect the fraud requires a more effective feature selection. Feature selection is a common technique in pre-processing which is used to reduce the dimensions of the data set. The current paper presents a hybrid approach to detect the fraud in credit cards. So, firstly, the primary features of the dataset are determined by the genetic algorithm followed by learning the model through hybridizing three different Type of neural network algorithms. Three mentioned algorithms are hybridized by the majority of votes, a weighing approach. In this approach the results and coefficients obtained by each neural network algorithm are voted to make the final output. The obtained results show that the mentioned technique, comparing the AFDM and AIS techniques, improves the cost and accuracy of prediction considerably. We have obtained an accuracy of 97.972 % by implementation of the aforementioned technique for this dataset. Similarly, the cost obtained through this study, has decreased almost 72% and 9% comparing with AIS and AFDM, respectively.