A parallel rough set based dependency calculation method for efficient feature selection

Feature selection is the process of selecting a subset of features without losing significant amount of useful information that is available in the original dataset. The use of rough set theory remains a prominent tool for this purpose. A number of techniques based on rough sets are based on the use of positive region based dependency measure which is a computationally expensive task. In this paper, we propose a novel dependency calculation technique termed as parallel dependency calculation technique. We propose to calculate dependency by directly finding the positive region based objects without calculating the positive region itself. As the objects are independent of each other so we propose to search these objects in parallel. The proposed technique was tested against the conventional dependency calculation technique and the results showed significant increase in performance, that is, overall 63.7% reduction in execution time and 96% reduction in required runtime memory was observed. In case of feature selection algorithms, there was 68% reduction in execution time when the proposed dependency calculation technique was used. The proposed method not only provides accuracy but is computationally less demanding than the conventional positive region based approach.