Removing noisy features via feature weights: preliminary results in mixed-model Gaussian distributions

In this article, we purpose an unsupervised feature selection algorithm that removes uniform noisy features encapsulated in the mixed model Gaussian distribution. The method is based on feature weighting principle and assumes that the noisy features have least feature weight and therefore have less or no contribution to cluster recovery. Experiments show that the proposed feature selection algorithm is more efficient in identifying noisy features compare to other similar algorithms like feature selection based on feature similarity (FSFS) or the intelligent K-Means feature selection(iKFS).

Date: 09/09/2016
Time: 16:00
Location: LB252

Share this post on: Twitter| Facebook| Google+| Email