Exploring Machine Learning Models for Fraud Identification Through Credit Cards in Financial Market

Credit card thieves prey on innocent, sociable people. Due to e-commerce and other websites, online payment methods have proliferated, which raises the likelihood of online fraud. Online transaction fraud has been analyzed and detected by professionals utilizing various machine-learning approaches as the fraud rates have been on the rise. This paper details a method for detecting financial fraud using recurrent neural networks on the Kaggle CCFD dataset, which takes into account the imbalance between classes and the dimensionality of features. Effective training and enhanced model performance were ensured by preprocessing the dataset using methods like PCA for feature reduction and SMOTE for data balance. The proposed RNN model demonstrated exceptional predictive capability, achieving an accuracy of 99.78%, precision of 99.73%, recall of 99.79%, and an F1score of 99.76%. These findings show that deep learning approaches outperform traditional ML models like Decision trees, SVM, and MLP when it comes to identifying fraudulent activities. These encouraging findings don't exclude the need for more optimization using methods like dropout regularization and hyperparameter tweaking; the model's performance might indicate overfitting.