Complementarity between the models indicates that ensemble techniques might be investigated to combine the results of several models in the most efficient way. The results obtained on manufacturing cost correspond to the statistical theory on comparative model performance. Last but not least, our results on a particular case study illustrate one of the epistemological debates about Data Mining and Machine Learning: is it possible to predict a phenomenon without understanding its physical foundations or underlying logic (Saporta, 2011)? If our analysis seems to point towards a positive answer, cost estimators should nonetheless rely on physics-based models as well as simple statistical models such as MLR to understand the main drivers of manufacturing cost.