Unsupervised Feature Selection Based on Rough Set Theory Using Direct Dependency Classes (DDC)

Abstract

The growth of data is considered to be at exponential rate in today’s advanced technological world where data is all around us in different forms. With the increase in number of records in a dataset there is increase seen in horizontal dimension also i.e. the no attributes are also being added up. When we perform data analysis or decision making activity, these attributes plays an important role. Sometimes, more the number of attributes vaguer the results are! Therefore, we have to select those attributes which contribute more in producing better results and leave those which are irrelevant. That is what we call dimensionality reduction or feature selection. Based on rough-set approach, direct dependency calculation algorithm is the primary procedure that reduces the no of attributes while preserving the key information. In this research study ,a direct dependency class calculation algorithm on unsupervised dataset is proposed, which has not been done before. The main goal is to extract useful features, reduce the code complexity and execution time while calculating dependencies of attributes on each other in a given dataset. This technique successfully performs feature selection by using two set of rules of direct dependency calculation. To verify the reduced execution time and algorithm complexity we carried out the experiment on standard datasets take from the UCI library. The results show great improvement in terms of feature selection accuracy and execution time with parallel processing. UDDC provides the accuracy above 95% when compared with other feature selection algorithms.