Applying i* in Conceptual Modelling in Machine Learning

Abstract

The i* framework is a popular and well-equipped technique for capturing the organizational environment and requirements of a system. However, i* heavily depends on the designer experience to cope with the idiosyncrasy of each specific field. While the machine learning field would benefit from a requirements representation, its complexity makes it unfeasible to directly use i*. The large number of constructs and nuances between elements puts a severe strain on the designer, leading to the creation of error-prone models. Therefore, in order to tackle this problem, we present an extension of i*. Our proposal covers the main gaps between machine learning and conceptual modeling with the aim of creating a suitable baseline methodology for machine learning requirements engineering. The advantage of our proposal is that our language specifies the main elements involved in machine learning models and constrains their interactions, filtering invalid designs and thus reducing the burden of knowledge while making the process less error-prone.