Smart decision support for invoice risk assessment
For her project, Lize investigates the use of A.I. and machine learning technology to assist operators in estimating the risk that financial documents carry; more specifically, the risk of buying an invoice from a party (seller), which concerns goods or services that were delivered to another party (debtor) and may or may not be paid in time. Estimating this risk is a time and data-intensive process that can be sped up by machine learning methods that are trained on large amounts of historical invoices. Besides classifiers and regressors, Lize investigated Learning-to-Rank methods for this task in her first journal paper, as they complement the graded nature of credit risk prediction. In order to deal with the uncertainty caused by the unknown outcomes of rejected invoices, classifiers with a reject option who abstain from making a prediction in case of uncertainty can be used. Lize focused on this topic in her DSAA conference paper. As a third topic, she is interested in building trust in predictions, which is necessary in case of sensitive (FinTech) data. She is therefore currently looking into local optimal explanation techniques.