If a company is to be acquired that is mainly based on a self-developed software system, as many tech companies today, the value of this software is a fundamental asset and therefore necessarily part of a due diligence process. It is assumed that the value corresponds to the effort a potential competitor would have, to develop the same software with the same functions. In software engineering, estimation models are frequently used to determine expected but yet unknown properties of software development processes or the developed systems, such as costs, times, number of developers, efforts, sizes, and complexities. The existing models have one or more problems: They are incomplete, incompatible or outdated. It is hard to compare and improve them as software technologies evolve quickly.

In this thesis an approach is developed with which models for mapping indicators of an outcome to the actual outcome can be easily trained and tested. While it has been developed for mapping indicators of software development costs and time to the actual costs and time, the approach is generally applicable to similar problems in other domains. Models should be developed by domain experts, not modelling experts, to provide domain understanding with the help of these models. As a result, the approach should be automated. Therefore, after training the models by optimization, predictions for unknown data points are calculated using the model. Models will be imprecise when trained at only few data points, hence, the approach is iterative allowing it to converge towards more and more precise models. In this thesis, an approach is developed with which indicators for mapping an estimated result to an actual result can be easily trained and tested.