A new MDL-based function for feature selection for Bayesian network classifiers
Mădălina M. Drugan, Linda C. van der Gaag · European Conference on Artificial Intelligence · 2004
Upon constructing a Bayesian network classifier from data, the accuracy of the resulting model can often be improved upon by selecting a subset of the available features. We show that the commonly used MDL function is not suited for feature selection. We introduce a new MDL-based function that is better tailored to this task. Our experimental results demonstrate that, with the new function, classifiers are yielded that have an accuracy comparable to the ones found with the MDL function, yet include fewer features.