A granular computing approach to data engineering

Granular computing is about computing with proper information granules for dealing with incomplete, uncertain or vague information. One of the main tasks in data engineering is concerning with data reduction. This paper presents an algorithm for data reduction based on a threshold derived from the concept of quality of approximation introduced in rough set theory. Experiments show that the improvement of prediction accuracies by data reduction is positively observable when the quality of approximation using reduced data set is at least 75% or its variation is small between raw and reduced data sets.