A Fast Wrapper Method for Feature Subset Selection

G. Richards, K. Brazier, and W. Wang (UK)


Data Mining, Feature subset selection.


This paper introduces a new feature subset selection algorithm, single-step selection. The algorithm has been tested on a range of datasets by constructing decision tree classifiers using all available features and comparing these with classifiers constructed using only a subset of features selected using the algorithm. On average, our feature subset selection algorithm not only significantly reduced the number of features (about 50% of features eliminated) but also improved the accuracy of the classifiers. We compared the performance of our method with another state of the art technique and determined that while the two methods resulted in similar improvements in classifier accuracy our method was determined to be simpler and very much faster.

Important Links:

Go Back