Supervised feature selection based on rough set theory and expectation-maximization algorithm

• Main Author: Zhang, Dong
• Other Authors: Wang Lipo
• Format: Final Year Project
• Language: English
• Published: 2013
• Subjects: DRNTU::Engineering
• Online Access: http://hdl.handle.net/10356/54382
• Institution: Nanyang Technological University

This report studies the feature selection based on the Expectation-Maximization Rough Set (RSEM) algorithm. The Expectation-Maximization clustering method extends the classical Rough Set concept of equivalent classes to tolerance classes, and enables the Feature Selection methods based on the traditional Rough Set theory to effectively deal with datasets with real values. The current RSEM algorithm is reviewed by both reproducing the results in the literature and applying three new classifiers to evaluate the features selected against a new fuzzy-rough algorithm. An improvement of the RSEM algorithm is proposed by changing the feature set evaluation method. The improved algorithm produces smaller feature sets by utilizing information hidden in the boundary region, without compromising the classification accuracies.