Abstract <p>This paper presents an innovative framework for enhancing credit fraud detection in banking by combining Autoencoder (AE), Long Short-Term Memory (LSTM) networks, and the Walrus Optimization Algorithm (WOA). The framework begins with the Autoencoder, which is utilized to pre-process high-dimensional transactional data. This step effectively reduces noise, extracts key features, and retains essential information, ensuring that the data fed into the model is of high quality. The LSTM network is then employed to capture the temporal dependencies and sequential patterns present in transaction data, enabling the model to detect fraudulent activities based on the order and timing of events. To further enhance performance, the Walrus Optimization Algorithm (WOA) is used to fine-tune the hyperparameters of the AE-LSTM model, optimizing both detection accuracy and computational efficiency. The integration of these techniques offers a comprehensive and scalable solution to the challenges of detecting fraudulent transactions in large, dynamic datasets typical in modern banking systems. By leveraging the strengths of each component—AE for data preprocessing, LSTM for sequential analysis, and WOA for optimization—the proposed model achieves significant improvements in fraud detection capabilities. This approach not only improves the detection of fraud in real-time banking transactions but also provides a more efficient and effective alternative to traditional fraud detection methods, addressing both accuracy and processing speed in a unified framework.</p> Graphical Abstract <p></p>