Predicting PC-SAFT pure-component parameters by machine learning using a molecular fingerprint as key input