https://www.researchgate.net/publication/360457994

https://www.researchgate.net/publication/347992937

https://www.researchgate.net/profile/Eugene-Lutsenko

Artificial Intelligence especially machine learning and deep learning-based techniques helps detect fraud, the models are trained using large datasets, which helps the models to learn the patterns related to fraud, and they show good performances, various research works have been performed recently on this topic.

The problem is that fraud is dynamic whereas machine learning models rely on historical training data to learn patterns. Kaggle labeled data sets incorrectly assume that all of the fraudulent transactions were identified.

So AI can be helpful in identifying previously identified modi operandi but fraudsters adapt their methods faster than companies can evolve detection methods.

Investigation is the bottleneck in most cases, not detection.