This is era of information. It is common to find the datasets with hundreds and thousands of features used by real world applications. Feature selection is a process to select subsets or features which are more informative. Feature selection technique is used to remove irrelevant and redundant features without losing much of the information. Recently Rough Set Theory (RST) becomes a dominant tool for FS. It is a theory which provides both data structures and methods to perform data analysis. Rough set theory has offered new ideas and trends for the features selection and deal with inconsistent information. Reduction of attribute is an important issue in rough set theory. Many feature selection techniques have been presented in literature using RST. However, majority of these techniques do not ensure optimal feature subsets and suffer serious performance bottlenecks especially in case of large datasets. In this thesis, we modified genetic algorithm to find subset of features within minimum execution time. In its conventional form, genetic algorithm is heuristic based approach; however, using genetic algorithm does not ensure the optimal feature sub selection. In this thesis, we have modified the algorithm such that the resulted feature subsets are not only the optimal but the resulting performance is also improved. The proposed approach was examined with other state of the art FS approaches various publically available datasets at UCI. Results show that efficiency and effectiveness of the proposed approach are better.