Photo by rawpixel on Unsplash
Research Question and Aim
The prediction of whether a customer will default on a loan is a long-standing challenge. In this paper, we explore both dealing with missing values in the data and data imbalances, as well as comparing various machine learning methods.
Conclusion
By adding weights to the algorithms we tested the limits of the two machine learning methods. By weighting the methods it is possible to achieve a 100% recall for both algorithms, albeit at the cost of accuracy. But for the prediction of loan defaults, such a trade-off is mostly worthwhile. And of the two algorithms, weighted logistic regression is the more impressive performer, yielded higher recall rate and AUC in both cases.
If interested in the details of the report, please click on the PDF block at the top of the page.
Yu Cao
I am passionate about leveraging large language models for multimodal learning, with a specific focus on unsupervised domain adaptation and domain generalization.