چكيده به لاتين
Identifying fraud and money laundering in the banking system, especially electronic banking, is very important due to the destructive effects on the economy of a country, and for this reason, banks and financial institutions use different methods to identify bank fraud. Recently, approaches based on artificial intelligence have been used a lot to detect bank violations, so in this research, a method based on artificial neural network and harmony search algorithm was used to detect fraud in bank transactions. The analyzed data included 1000 transactions (508 healthy transactions and 492 suspicious transactions). Each data is defined by 30 attributes. The decision tree model, artificial neural network and component analysis (PCA) were used to identify the type of transaction. At first, the mentioned models were trained with the entire dataset. Then, using the PCA method, the optimal features were selected and the decision tree and artificial neural network models were trained again to recognize the type of transaction. Finally, the artificial neural network model was optimized by the harmony search algorithm. The results of detecting suspicious transactions by the entire set of features showed that the overall accuracy of the decision tree in the transaction type classification is equal to 93.80% and the final accuracy of the artificial neural network with trainscg, trainbr and trainlm functions is equal to 94.3%, 2.2 It was 99% and 96.6%. Also, the final accuracy of PCA-ANN-trainbr and PCA-DT models in the classification of transaction type with selected data by PCA was 99.3% and 94.2%. Finally, the overall accuracy of the HSA-ANN model in the transaction type classification was 99.50%, and the sensitivity of this method in detecting healthy and suspicious transactions was 100% and 99%, respectively. In general, the proposed research was more accurate than other similar researches, so that it improved the results of transaction type classification accuracy between 0.5% and 11%.
