This paper addresses the challenge of catastrophic forgetting in continual credit card fraud detection. The study evaluates various strategies and models to balance plasticity (learning new patterns) and stability (remembering old patterns) in real-time fraud detection systems. The authors propose a plasticity/stability visualization matrix to quantify the forgetting phenomenon in data streams with delayed feedback. They compare six strategies and 13 different models on a real-world dataset of over 50 million e-commerce transactions. The results show that ensemble methods, such as FrzE (frozen neural networks), perform best in terms of accuracy and reduced forgetting. The study highlights the importance of balancing plasticity and stability in fraud detection systems, especially given the dynamic nature of fraud patterns. The paper also discusses the trade-offs between different approaches and emphasizes the need for adaptive strategies to handle non-stationary data. The findings contribute to the field of continual learning by providing a practical framework for assessing and mitigating catastrophic forgetting in real-world applications.This paper addresses the challenge of catastrophic forgetting in continual credit card fraud detection. The study evaluates various strategies and models to balance plasticity (learning new patterns) and stability (remembering old patterns) in real-time fraud detection systems. The authors propose a plasticity/stability visualization matrix to quantify the forgetting phenomenon in data streams with delayed feedback. They compare six strategies and 13 different models on a real-world dataset of over 50 million e-commerce transactions. The results show that ensemble methods, such as FrzE (frozen neural networks), perform best in terms of accuracy and reduced forgetting. The study highlights the importance of balancing plasticity and stability in fraud detection systems, especially given the dynamic nature of fraud patterns. The paper also discusses the trade-offs between different approaches and emphasizes the need for adaptive strategies to handle non-stationary data. The findings contribute to the field of continual learning by providing a practical framework for assessing and mitigating catastrophic forgetting in real-world applications.