We have developed a new approach (MDLEP) to learning Bayesian network structures based on the Minimum Description Length (MDL) principle and Evolutionary Programming (EP). It integrates a MDL metric founded on information theory and several new genetic operators including structure-guided operators, a knowledge-guided operator, a freeze operator, and a defrost operator for the discovery process. In contrast, existing techniques based on genetic algorithms (GA) only adopt classical genetic operators. We conduct a series of experiments to demonstrate the performance of our approach and to compare it with that of the GA approach developed in a recent work. The empirical results illustrate that our approach is superior both in terms of quality of the solutions and computational time for data sets we have tested. In particular, our approach can scale up and discover extremely good networks from large benchmark data sets of 10000 cases. Lastly, our MDLEP approach does not need to impose the restriction of having a complete variable ordering as input.
|Title of host publication||Genetic Programming 1998: Proceedings of the Third Annual Conference|
|Publication status||Published - Jul 1998|
|Event||The Third Genetic Programming Conference 1998 - University of Wisconsin, Madison, Wisconsin, United States|
Duration: 22 Jul 1998 → 25 Jul 1998
|Conference||The Third Genetic Programming Conference 1998|
|Period||22/07/98 → 25/07/98|
LAM, W., WONG, M. L., LEUNG, K. S., & NGAN, P. S. (1998). Discover Probabilistic Knowledge from Databases Using Evolutionary Computation and Minimum Description Length Principle. In Genetic Programming 1998: Proceedings of the Third Annual Conference (pp. 786-794)