Asian Research Thesis Index

Abstract

Volumetric increase in data along with the curse of dimensionality has diverted the recent trends of computer science. Processing such a massive amount of data is a computationally expensive job. Feature selection is the process of selecting subset of data from the entire dataset that contains most of the information. The selected subset is called Reduct. Feature selection has materialized the idea of jumbling with attributes. Subset of attributes is favored which bounces the same information as the wide-ranging set of variables. Various dynamic reduct finding algorithms have been proposed. Dynamic reducts is an extension to the idea of reduct extraction based on rough set. Sub-tables are randomly drawn from the original decision table and reducts are extracted from these sub-tables. These reducts are considered to be the stable reducts for complete dataset. However, all the existing dynamic reduct finding algorithms are computationally too expensive to be used for datasets beyond smaller size. In this research, a novel dynamic reduct finding technique based on rough set theory is proposed, where dynamic reducts and relative dependency are the two key notions. Reducts are selected, optimized and further generalized through strenuous Parallel Feature Sampling (PFS) algorithm. In-depth analysis is performed using various benchmark datasets to justify the proposed approach. Results have shown that the proposed algorithm outperforms the existing state of the art approaches in terms of both efficiency and effectiveness.

Add/Update Thesis