Fast feature selection and ranking system

Abstract

In the real world, especially the so called “Information Age”, we are nearly buried in the world of data and information. Consequently, feature selection has successfully become an important and necessary technique in helping us looking for the domain knowledge and underlying concept from datasets that are usually of enormous dimensionality comprised of many features and thousands of instances, such as the transactional dataset. During the recent years, different feature selection methods with various approaches, mainly based on search strategies and evaluation criteria, has been proposed. Interestingly, one of the most common and major problem they have encountered is the overall performance of the algorithm deployed, especially when the datasets are getting larger and larger. In this project, it aims to compare different approaches used for feature selection and at last an advanced method, called “Quick-EB”, with significantly improved efficiency and performance, will be introduced and demonstrated on some real-life applications to find out the most important features efficiently and effectively which are ranked based on the entropy measurement.