Abstract
Neighborhood rough set is a powerful tool to deal with continuous value information systems. Graphics processing unit (GPU) computing can efficiently accelerate the calculation of the attribute reduction and approximation sets based on matrix. In this paper, were write neighborhood approximation sets in the matrix-based form. Based on the matrix-based neighborhood approximation sets, we propose the relative dependency degree of attributes and the corresponding algorithm (DBM). Furthermore, we design the reduction algorithm (ARNI) for continuous value information systems. Compared with other algorithms, ARNI can effectively remove redundant attributes, and less affect the classification accuracy. On the other hand, the experiment shows ARNI based on the matrixing rough set model can significantly speed up by GPU. The speedup is many times over the central processing unit implementation.
Highlights
The explosive growth of data volume increases the complexity of data, and makes data processing more difficult than before
We proposed the relative dependency degree of attributes based on matrixing neighborhood approximation sets, and the corresponding reduction algorithms (DBM and Attribute Reduction based on Neighborhood Importance Degree (ARNI))
Based on the previous propositions and definitions, we design two algorithms Dependency Degree based on Boolean Matrix (DBM) and Attribute Reduction based on Neighborhood Importance Degree (ARNI)
Summary
The explosive growth of data volume increases the complexity of data, and makes data processing more difficult than before. Discretizing the continuous values exists some uncertainty and may lose some essential information To solve this problem, many rough set models have been proposed, such as fuzzy rough sets [3,4,5,6], covering rough sets [7,8,9], semimonolayer cover rough set [10], neighborhood rough sets [11,12,13,14], granule-based rough sets [15,16,17]. Zhang et al [30] adopted a multi-GPU solution to accelerate their algorithm about a parallel method for computing approximations based on matrix and achieved 334.9 times of acceleration compared to the CPU. We proposed the relative dependency degree of attributes based on matrixing neighborhood approximation sets, and the corresponding reduction algorithms (DBM and ARNI).
Talk to us
Join us for a 30 min session where you can share your feedback and ask us any queries you have