NEC: A nested equivalence class-based dependency calculation approach for fast feature selection using rough set theory

Abstract Feature selection plays an important role in data mining and machine learning tasks. As one of the most effective methods for feature selection, rough set theory provides a systematic theoretical framework for consistency-based feature selection, in which positive region-based dependency calculation is the most important step. However, it is time-consuming, and although many improved algorithms have been proposed, they are still computationally time-consuming. Therefore, to overcome this shortcoming, in this study, a nested equivalence class (NEC) approach is introduced to calculate dependency. The proposed method starts from the finest partition of the universe, and then extracts and uses the known knowledge of reducts in a decision table to construct an NEC. The proposed method not only simplifies dependency calculation but also reduces the universe correspondingly, in most cases. Using the proposed NEC-based approach, a number of representative heuristic- and swarm intelligence-based feature selection algorithms that apply rough set theory were enhanced. Note that the feature subset selected by each modified algorithm and that selected by the original algorithm were the same. Experiments conducted using 33 datasets from the UCI repository and KDD Cup competition, which included large-scale and high-dimensional datasets, demonstrated the efficiency and effectiveness of the proposed method.