An unsupervised feature selection framework based on clustering
PAKDD'11 Proceedings of the 15th international conference on New Frontiers in Applied Data Mining
| Hi-index | 0.01 |
Many algorithms for data mining and machine learning can only process discrete attributes. In order to use these algorithms when some attributes are numeric, the numeric attributes must be discretized. Because of the prevalent of normal distribution, an approximate equal frequency discretization method based on normal distribution is presented. The method is simple to implement. Computing complexity of this method is nearly linear with the size of dataset and can be applied to large size dataset. We compare this method with some other discretization methods on the UCI datasets. The experiment result shows that this unsupervised discretization method is effective and practicable