Identifying fraud in the banking industry stands as an ongoing challenge, owing to the shifting patterns of fraudulent conduct and the variety of large-scale transactions. Classic machine learning algorithms frequently struggle with massive data structures and fail to grasp the temporal correlations required for fraud detection. To cope with such limitations, this study supplies an architecture that combines Quantum Feature Engineering with Quantum Long Short-Term Memory (QLSTM). The approach improves feature representation by mapping transaction traits to quantum states via Angle Encoding, whereas the QLSTM model combines quantum computational capability with temporal sequence modeling. The suggested architecture considerably enhances fraud detection performance, establishing a scalable, reliable solution for the banking industry.