A New Modified Approach to Rough Set Feature Selection

Extracting useful information from a huge data collection is an important and challenging issue. Feature selection (FS) refers to the problem of selecting minimal relevant features which produce the most predictive outcome and retaining the original meaning of the features after reduction. One of the successful techniques for feature selection from datasets is the rough set theory (RST). This paper starts with an outline of the fundamental concepts behind Rough set related to FS methods. Feature selection algorithms like Quickreduct (QR), Relative Reduct (RR), presented here. One new modified method based on Lower approximation base feature selection i.e. Improved Quickreduct algorithm is proposed which selects minimal subset of features and execution time is better than the original methods. This is performed using a stopping criterion with a threshold and with the concept of significance of features. A comparative study of the algorithms and the experiments are carried out on eight public domain datasets available in UCI machine learning repository to analyze the performance study.