An efficient stripped cover-based accelerator for reduction of attributes in incomplete decision tables

Abstract Attribute reduction in incomplete decision tables plays an extremely important role in machine learning, data mining, pattern recognition, especially for experts and intelligent systems. Many different reducts have been given in the rough set approach to find a promising reduct. However, efficiently extracting a reduct from large-scale incomplete data sets is time-consuming and becomes a challenging research problem. Although researchers have spent a lot of efforts for improving the computational efficiency, most existing methods have quite high complexity and focus mostly on the positive region reduct. To accelerate the attribute reduction process, we firstly introduce in this paper a new concept of stripped covers. Then, we investigate vital properties of stripped covers as well as provide attribute significance measures. By using these measures, we propose an effective and efficient heuristic algorithm framework for fast computation of popular reduct types. It is also worthwhile to mention that our algorithm has better time complexity than existing methods. Furthermore, the performance of the proposed method is experimentally demonstrated across multiple real-world datasets and compared with the main state-of-the-art methods. The results showed that our method outperforms the compared methods in the terms of obtained reduct size, computational time and classification accuracy.