Real-Time Synthetic Identity Fraud Detection Using Recurrent Neural Networks for Sequential Data Analysis

The demand for accurate time-series classification has increased in recent years, but common models, such as GBM and LSTM, have been struggling with high dimensionality and nonlinear patterns in recent times. Therefore, this paper proposes an RNN architecture that can extract temporal dependency more efficiently. Dropout techniques reduce the overfitting process, while optimized training helps in much faster convergence. In addition, the performance evaluation using accuracy, precision, recall, and F1 score verifies that the proposed RNN outperformed conventional models by yielding 99.21 % for accuracy, 98.73 % for precision, 99.53 % for recall, and 99.13 % for F 1. These results prove the robustness and scalability of the model on datasets of variations. The model has wider implications for advanced RNNs in predictive analytics and lays the basic foundation for further research related to time-series classification.

Перевод пока недоступен