This study uses decision tree and random forest regressions to improve the accuracy of an approximation of credit default swap (CDS) spreads called the Equity-to-Credit (E2C) formula based on a time span including the COVID-19 pandemic period. Certain sections are dedicated to explaining deeper important concepts in machine learning. Random forest regressions run with the E2C and selected additional financial data results in an accuracy in CDS approximations of 82% out-of-sample. The transparency property of these algorithms confirms that, for CDS spreads’ forecasting, the most used feature is the E2C formula and to a lower extent companies’ debt rating and size.