A Null Value Estimation Method Based on Similarity Predictions in Rough Sets

Abstract

In this paper, we utilize rough set theory as a tool to deal with the problem of null-value estimation in an incomplete information system, a rating mechanism in collaborative filtering technology is introduced into this paper for the weakness of null value estimation based on similar relational algorithm (SIM-EM), such as no sparse degree process and low accuracy, and an improved null value estimation method, which based on SIM-EM is proposed from the perspective of similarity. The null value data is predicted and filled through the similarity of objects; otherwise a dual feature weight method is proposed according to the attribute's feature in rough set, which improves the accuracy in similarity calculation; the improved algorithm is good at dealing with sparse rough set, and the accuracy and the mean absolute error is better than the original method.