Machine Learning techniques can be used to improve the performance of intelligent software systems. The performance of any Machine Learning algorithm mainly depends on the quality and relevance of the training data. But, in real world the data is noisy, uncertain and often characterized by a number of features. Existence of uncertainties and the presence of irrelevant features in the high dimensional datasets often degrade the performance of the machine learning algorithms in all aspects. In this paper, the concepts of Rough Set Theory(RST) are applied to remove inconsistencies in data and various RST based algorithms are applied to identify the most prominent features in the data. The performance of the RST based reduct generation algorithms are compared by submitting the feature subsets to the RST based decision tree classification. Experiments were conducted on some of the medical datasets taken from Irvine UCI machine learning repository.