New Feature Engineering Framework for Deep Learning in Financial Fraud Detection

The total losses through online banking in the United Kingdom have increased because fraudulent techniques progressed and used advanced technology. Using history transaction data is limit for discovering various patterns of fraudsters. Autoencoder has a high possibility to discover action without considering unbalanced fraud class data. Although autoencoder model uses only majority data, our hypothesis, if original itself feature vectors related transactions before inputting then performance detection improved. A new engineering framework built that can create select effective features deep learning remote detection. Based on proposed [19], been created using methods based their importance. In experiment, real-life dataset which was provided by private bank Europe models with three different types datasets: With selected features. We also adjusted threshold values (1 4) evaluated them datasets. result demonstrates are significantly improved than ones