Testing is one of the most important means of quality assurance for systems. Using test technologies at an early stage enables developers to find errors faster during the development process and save on development costs. In fuzz testing, testers confront the program with a large amount of random data in an attempt to cause it to crash. This way, developers can discover security gaps in the software.

The paper describes an approach for identifying and regenerating properties of input data to execute specific program parts in a targeted manner. For this purpose, pattern mining is performed on models that represent the input data using trees. The pattern mining is intended to identify similarities in input data that are necessary for the execution of targeted program parts. The properties are generated by inserting learned subtrees into new inputs. The presented approach aims to increase the coverage of the program code during testing and to identify additional security vulnerabilities.