Explainable machine learning in identifying credit card defaulters

Abstract

Machine learning is fast becoming one of the central solutions to various real-world problems. Thanks to powerful hardware and large datasets, training a machine learning model has become easier and more rewarding. However, an inherent problem in various machine learning models is a lack of understanding of what goes on ’under the hood’. A lack of explainability and interpretability leads to lower levels of trust in the model's predictions, which means it can't be used in sensitive applications like diagnosing medical ailments and detecting terrorism. This has led to various advances in making machine learning explainable. In this paper various black-box models are used to classify credit card defaulters. These models are compared using different performance metrics, and explanations of these models are provided using a model-agnostic explainer. Finally, the best model-explainer combo is proposed with potential areas of future exploration.