SMOTE-AAE-RF: A Novel Approach to Identifying Financial Fraud in Imbalanced Data Circumstances

Nowadays, artificial intelligence (AI) and deep learning-based fraud detection methods are more powerful than conventional machine learning methods. However, the financial sector, including banks, insurance companies, the share market, and non-banking financial companies (NBFCs), also widely adopted these techniques in India. But these AI and deep learning techniques still face several challenges, including class-imbalance issues, class overlapping, noise, and high computational costs. In this paper, we have analyzed these challenges in data sets in terms of data complexity. Reducing the data complexity of the dataset will undoubtedly enhance the performance of the model. We also suggest a novel SMOTE-AAE-RF technique to detect fraud in the financial domain. This technique combines the Synthetic Minority Oversampling Technique (SMOTE), the Adversarial Autoencoder (AAE), and the Random Forest (RF). When the data set is skewed, we use the SMOTE technique to resample it. An adversarial autoencoder is used as a filter to filter out the samples that were generated by the SMOTE technique, and the RF classifier is used to detect whether the transactions are genuine or fraudulent. We analyzed the superiority of the suggested method on various performance measures, and the results indicate that it outperformed its competitor methods.