The fraud in banking has increased considerably due to the increasing use of digital transactions as well as internet banking. Conventional fraud detection systems are unable to keep up with the changing trends and most of them are unable to detect fraud in time. The deep learning models have gained popularity as a potentially competent option because they can learn complicated patterns from massive transactional data. But they are also of a black-box nature, which hinders transparency and trust, especially in critical sectors such as banking. To address this shortcoming, XAI is being progressively added to fraud detection systems to make sure the decisions made by deep learning models are understandable to stakeholders. This paper describes the current state of the use of deep learning in fraud detection in the banking industry and how it can be augmented using XAI techniques like SHAP, LIME, and attention mechanisms to improve the reliability, interpretability, and efficacy of the resulting fraud detection systems. It is the first survey that summarizes publicly available datasets like Kaggle Credit Card Fraud Detection dataset and the IEEE-CIS Fraud Detection dataset, and compares deep learning models like CNNs, RNNs, LSTMs, Autoencoders, and GNNs. The key metrics through which these models are compared include Accuracy, Precision, Recall, F1-score and AUC-ROC. The uniqueness of this work lies in coupling deep learning techniques with XAI techniques (SHAP and LIME) to offer fraud detection that is transparent and friendly to regulators. It also takes a census of various deep learning models such as CNNs, RNNs, LSTMs, Autoencoders, and GNNs in transaction anomaly detection. Moreover, the paper also identifies existing datasets, performance benchmarks, issues to be addressed like data imbalance and adversarial fraud, and a future roadmap. The statistical data, performance charts, and model comparison bar graphs will be incorporated as they will give visual evidence for the findings. The paper will attempt to close the gap between the accuracy and interpretability of AI, and consequently, ensure the responsible use of AI in the banking sector.

Fraud Detection Banking Sector Deep Learning Explainable AI (XAI) Digital Transactions Internet Banking SHAP LIME Attention Mechanisms Convolutional Neural Networks (CNNs) Recurrent Neural Networks (RNNs) Long Short-Term Memory (LSTM)