Towards machine learning based configuration

Nowadays, configuration technology is one of the most well-known and utilized applications of Artificial Intelligence (AI) which relies mainly on constraint-based approaches. Dependencies between features of the configured product are modeled as constraint satisfaction problems (CSP). This approach inherits some drawbacks considering the huge effort knowledge engineers have in maintaining knowledge bases, especially in complex configuration scenarios. In this paper, we propose an alternative configuration approach by utilizing machine learning (ML) algorithms and show that this technology might be a gamechanger for future configuration and recommendation approaches. To demonstrate the possibilities of ML in the configuration domain we implemented a prototype and showed its effectiveness in a short case study.