To satisfy customer needs in the best way, companies offer them an almost infinite number of product variants. Although, an identical product was not built before, the values of its attributes must be determined during the product configuration process. This paper introduces a methodical approach to predict the values of product attributes based on customer feature configurations using machine learning. Machine learning reduces the effort compared to rule-based expert systems and is both, more accurate and faster. The approach was validated by predicting vehicle weights using industrial data.

Companies are increasingly struggling to manage their complex product portfolios. Since they do not fully understand the complexity, intelligent solutions are required. Emerging technologies and tools offer new ways to deal with existing problems. With the help of clustering, similarities between product variants can be identified automatically, and complexity can be systematically reduced. This article aims to develop a methodological approach to identify correlations between product variants in complex product portfolios automatically by using clustering algorithms. The approach includes the systematic cleaning and transformation of product portfolio data. In addition, a guide for algorithm selection and evaluation of clustering results is presented. As the last step, the results are systematically analysed and visualised. To validate the methodical approach, it is applied to a real-world data set of a commercial vehicle manufacturer and the usefulness of the results is confirmed in an expert workshop.