| Abstract [eng] |
The choice of deferring payment i.e. granting trade credit to customers has a significant impact on both long-term relationship between companies and the working capital management chain of both companies. Because up to 25 percent of short-term assets may consist of such trade credit, the granting of this credit must be controlled, and each customer must be assessed both before the first invoice is issued and throughout further cooperation. Such assessment of the buyer's creditworthiness may be made based on information obtained from external sources, but this information can be updated semi-annually and must be is paid for. Meanwhile, the customer's solvency history is accumulated within the company itself, which can be used to assess the buyer' s current situation and update his creditworthiness assessment at the current moment. The aim of the work is to apply machine learning methods to assess the creditworthiness of companies' businesses. First, create a tool for clustering buyers and refine expert credit assessment. Using this information, new buyers or buyers with changed solvency data would be automatically classified into different levels of creditworthiness. During the study, buyers were clustered based on their 2018 solvency history using the K-mean method. 38,99 percent of the cases were revised with an expert creditworthiness assessment. Then, machine learning was performed using Support vector with radial basis function and Random forest methods and their results were compared and was found that the latter can correctly assess creditworthiness up to 90 percent of new buyers with information only on their solvency. After examining the dependence of the accuracy of the classification results by the variables, six of them were selected, which were considered to have the most significant influence in the accuracy of the prediction. |
