In the course of increasing individualization of customer demand, configurable products are gaining importance. Nowadays, variant-specific bills of materials and routings for configurable products are created with the help of rule-based configuration systems, so-called low-level configuration systems. The rules and generic structures on which such configuration systems are based are created manually today. This is challenging because it can be difficult and sometimes impossible to directly transfer expert knowledge into those systems. Furthermore documents that have already been created by experts in the past such as bills of material and routings contain relevant information as well which may be exploited to compose configuration systems. However, in the literature, there are no approaches yet to systematically transfer expert knowledge into configuration systems or to consider existing documents. In addition, the creation of such configuration systems is prone to error due to their complexity. Although there are already numerous approaches to the formal testing of configuration systems, approaches based on data analysis to support the validation of such systems have not yet been considered. Therefore, in this paper an approach is presented to automatically create low-level configuration systems by means of exemplary variant-specific bill of materials and routings using machine learning. The super bill of materials and the super routing as well as the dependencies between the product characteristics and the components respectively the operations are learned. Furthermore, it is shown how errors in the input data as well as errors in the resulting low-level configuration system can be detected by means of anomaly detection.