Enhancing Credit Card Default Prediction: Prioritizing Recall Over Accuracy

Given the increase in credit card usage, it is crucial to predict possible defaults to effectively handle credit risk. However, a lot of prediction models ignore the effects of false negatives in favour of narrowly focusing on maximizing overall accuracy. Credit default model is one of the most useful analytical models that is often used for risk assessment. Therefore, our goal is to ensure that we improve its recall yet preserve precision and overall accuracy. Many machine learning methods such as feature selection, ensembling and sampling were used on a Taiwanese credit card dataset. The optimized SVM model achieved a recall of 51.47% while balancing recall and precision. The report shows how other limiting factors can be avoided thus minimizing false negative cases that may lead to lenders caught off guard with huge financial repercussions when unexpected loans go into default. The conclusion demonstrates the fact that revising the credit card default prediction paradigm should be done by prioritizing recall in addition to relying on the level of overall accuracy. This finding does not just fill in a gap in the field or study of risk assessment, but also in its practical use. His balanced method of modelling provides the sector a practical way to improve credit default forecasts.