[Current approaches and challenges in addressing class imbalance in medical prediction models]

With the rise of personalized medicine and the rapid development of big data technology, medical prediction models have become increasingly important in disease diagnosis, prognosis assessment, and risk stratification. However, class imbalance is a common problem in medical data, which can result in models being overly trained toward the majority class rather than the minority class, influencing the detection power and clinical application value. This paper systematically summarizes traditional methods in addressing class imbalance, including data pre-processing and algorithm level strategies, and introduces the applications of new technologies such as generative adversarial networks and transfer learning and suggests key considerations and potential research focus for addressing class imbalance to provide reference for researchers to select appropriate strategies.