Estimating function points through conceptual models

The estimation of software has been an issue in the last decades. Being able to estimate the size and, therefore, the cost in the early stages of the software life cycle has always been an added value of a project. Since the end of the 70s, the function points have been the most used metric for measuring the functional size of an application, and recently, the European Parliament confirmed their importance. The function points have the problem that are not objective because they are estimated starting from natural language, and every analyst can write the requirements differently. Using more formal models, e.g., conceptual model, as input for the function points analysis can mitigate this issue, opening the possibility to automate such an activity. This thesis has the two-fold objective of conducting i) a SLR (Systematic Literature Review) and ii) exploring and comparing different techniques for estimating function points starting from conceptual models. The SLR has the goal to discover if and how this approach has been explored in the past and no similar studies exist in the literature. The results of SLR show that this theme has been discussed little; in particular, only some studies address the problem of estimating function points starting from conceptual models. Carrying out a study that collects and compares different approaches using conceptual models is a novel contribution to the field of Function Point estimation. Using the results of the SLR, several techniques for estimating function points starting from conceptual models have been analyzed, using ER (Entity-Relationship) and BPMN (Business Process Modelling Notation). These techniques include multiple linear regression, neural networks, and ChatGPT. Results are promising, but a larger dataset is still needed to validate the first two techniques better.