A Local Density Approach for Unsupervised Feature Discretization

Discretization is an important preprocess in data mining tasks. Considering the density distribution of attributes, this paper proposes a novel discretization approach. The time complexity is O (m *n * logn ) as EW and PKID, so it can scale to large datasets. We use the datasets from the UCI repository to perform the experiments and compare the effects with some current discretization methods; the experimental results demonstrate that our method is effective and practicable.