Title |
Uncertain knowledge and machine learning application for trade risk prediction and sensitivity identification / |
Authors |
Alzbutas, Robertas ; DundulienÄ—, Ieva |
Full Text |
|
Is Part of |
HiTEc COST Action: Text, functional and other high-dimesional data in econometrics: new models, methods, applications: working groups meetings, Limassol, Cyprus, 3-4 April 2023.. Limassol : [s.n]. 2023, A0187, p. 1 |
Keywords [eng] |
trade risk ; machine learning ; sensitivity identification |
Abstract [eng] |
Trade risk prediction can benefit from integrating expert information, big data methods, and artificial intelligence. Companies seek automated solutions to reduce bad debt probability and the default risk of not getting back trade credit. The focus could be on a framework of expert knowledge incorporation into the machine learning pipeline. Various prediction methods are considered, including specialist knowledge elicitation and ensemble model applications. Uncertain knowledge about risk is integrated into the review process of unsupervised learning and clustering results as further classification based on the risk expert recommendations. Classes confirmed by the expert are used as labels for following supervised machine learning and risk prediction. The framework was tested with a real-world dataset, and the risk prediction correctly identified about 90\% of the cases. Feasibility of probabilistic uncertainty and sensitivity analysis using sample and variance-based methods were also considered for risk sensitivity to uncertain features. Sampling-based methods, tolerance intervals, and the Wilks method could be applicable for computationally intensive modelling to speed up the classification training and updating process. This additionally helps timely incorporate relevant data and, in advance, better identify potentially risky companies, avoid risky transactions, and reduce the risk of loss. |
Published |
Limassol : [s.n] |
Type |
Conference paper |
Language |
English |
Publication date |
2023 |