The rapidly growing field of data mining has the potential of improving performance of existing scheduling systems. Such systems generate large amounts of data, which is often not utilized to its potential. The problem is whether it is possible to discover the implicit knowledge behind scheduling practice and then, with this knowledge, we could improve current scheduling practice. In this dissertation, we propose a novel methodology for generating scheduling rules using a data-driven approach. We show how to use data mining to discover previously unknown dispatching rules by applying the learning algorithms directly to production data. We also consider how by using this new approach unexpected knowledge and insights can be obtained, in a manner that would not be possible if an explicit model of the system or the basic scheduling rules had to be obtained beforehand. However, direct data mining of production data can at least mimic scheduling practices. The problem is whether scheduling practice could be improved with the knowledge discovered by data mining. We propose to combine data mining with optimization for effective production. In this approach, we use a genetic algorithm to find a heuristic solution to the optimal instances selection problem, and then induce a decision tree from this subset of instances. The optimal instance selection can be viewed as determining the best practices from what has been done in the past, and the data mining can then learn new dispatching rules from those best practices.