Evaluation of Machine and Deep Learning Methods for Fraud Detection

The increase in fraudulent activities seriously threatens financial reliability and highlights the importance of detection and prevention solutions. Due to the complex and developing nature of fraudulent activities, new and advanced solutions are required to detect these activities such as machine learning (ML) and deep learning (DL) methods. In this scope, three different datasets were used where each of them was tested with three different methods for class imbalance across three different ML models and six different DL models. Their performance was measured with four different types of metrics. The current study provided valuable insights into the performance of ML and DL models on highly imbalanced datasets. These findings are not only relevant for fraud detection but also hold significant implications for sectors such as disease diagnosis, cybersecurity, insurance, and image recognition, where imbalanced datasets are also prevalent.