Gaining the empathy and trust of customers is paramount in the financial domain. However, the recurring occurrence of fraudulent activities undermines both of these factors. ATM fraud is a prevalent issue faced in today’s banking landscape. The critical challenges in fraud datasets are highly imbalanced datasets, evolving fraud patterns, and lack of explainability. In this study, we handled these techniques on an ATM transaction dataset collected from India. In binary classification, we investigated the effectiveness of various oversampling techniques, such as the Synthetic Minority Oversampling Technique (SMOTE) and its variants, Generative Adversarial Networks (GAN), to achieve oversampling. Gradient Boosting Tree (GBT), outperformed the rest of the techniques by achieving an AUC of 0.963, and Decision Tree (DT) stands second with an AUC of 0.958. In terms of complexity and interpretability, DT is the winner. Among the oversampling approaches, SMOTE and its variants performed better. We incorporated explainable artificial intelligence (XAI) and Causal Inference (CI) in the fraud detection framework and studied them via various analyses. Further, we provided managerial impact.