A machine learning approach to estimate product costs in the early product design phase: a use case from the automotive industry

Abstract

The management and controlling of product costs play an important role in the early stages of value creation, since 70 - 80 % final product costs are determined in product development and design processes. Traditionally, the focus in business practice is often on the organizational controlling unit that records the corresponding costs when they occur. In those, later phases of the product lifecycle, the possibilities of exerting influence are very limited. Consequently, an accurate estimation of costs in the early design phase is an important instrument to support strategic decisions in the product engineering process. An intelligent approach based on machine learning to support cost management in the early design stage is presented. In a real-world case study using wheel cost data from a large original equipment manufacturer (OEM) we investigate the performance of a cross-validated machine learning framework using a greedy sequential forward feature selection method for dimensionality reduction, as well as grid search hyperparameter tuning in order to compare six regression algorithms. The results of the study show that it is possible to predict the costs of product components in the early design phase with an R2 of 0.960 using only seven features coming from the product design and development department. By a cluster analysis and scatter plots of prediction accuracies it is shown where the deviations in cost estimation come from. The study supports the statement that machine learning models are a promising instrument for product managers and cost engineers.