A COMPARATIVE STUDY OF VARIOUS ROUGH SET THEORY ALGORITHMS FOR FEATURE SELECTIO N

Now a days, Machine Learning[1] are being applied successfully to improve the performance of many intelligent systems like Weather forecasting, Face detection, Image Classification, Disease diagnosis, Speech recognition, Signal denoising e.t.c., Machine Le arning techniques help in developing an efficient intelligent system without much intervention of humans. Decision Tree Classification[2] is widely used in machine learning for classification. The primary factor that affects the performance of any machin e learning algorithm is the quality of training data. Before modeling a classification technique, the quality of the data must be checked. Secondly, the dimensionality of the training data also affects the computational complexity of the machine learning a lgorithm. As data is characterized by many features and not all these features contribute for a particular task and hence there is a great demand to identify the features or attributes that are relevant for a particular task to reduce the feature space so as to reduce the computational complexity of the learning model. In this paper, the concepts of the most popular Rough Set Theory(RST)[3] is applied for inconsistent removal, feature subset selection and also to induce decision tree. At first, the basic co ncepts of RST are used to identify and eliminate inconsistencies in the data and then various versions of RST based Quick Reduct algorithm[4] are applied to know the most relevant set of features in the training data. And to compare the effectiveness of t he RST based reduct generation algorithms, the generated feature subsets are submitted to the RST based decision tree classification and the obtained prediction accuracies are compared.