Algorithms for combinatorial problems, e.g. production scheduling, often requires significant specialized knowledge. Therefore, it is usually not possible to take an algorithm that works well on a certain type of instances and use it on instances having a different character. In this case, it would be advantageous to have a mechanism able to analyze the new instances and adopt the behavior of the combinatorial algorithm, i.e. a method that is known as the data-driven approach. The aim of this project is to take a simple combinatorial problem, e.g. a simple production scheduling problem, and design a machine learning based algorithm able to solve it.