Article Preview
Top1. Introduction
With the advancement of technologies nowadays, more data are created every day, leaving analysts to determine the popular trends and activities. New data warehousing and mining methods are designed to deal with that issue. Recent approaches include Scalable Biclustering (Balamane, 2021), Filter-Wrapper Incremental Algorithms (Giang et al., 2020 & 2021), Spatiotemporal mining (Monteiro et al., 2021), Deep Learning based on GAN (Li et al., 2021), Social Mining (Jung and Chung, 2021), Clustering for Big Data (Kwok et al., 2002) (Purandhar et al., 2021), Image segmentation (Jaha et al., 2019), Enhancing quality of signals (Jaha et al., 2019), Recommendation Systems (Patro et al., 2020), Decision-making (Quek et al., 2019), Exoplanet detection (Priyadarshini et al., 2021) and Predicting (Earth Pritam et al., 2019). Among those trends, attribute reduction is one of the elementary problems in Data Mining frequently used in those methods. This problem aims to determine a core number of attributes or the reducts from the original set of attributes to achieve the best classification accuracy. The initiative of this problem is that not all attributes in the original set have the same impact on the decision variables. In other words, some attributes are meaningful, and it is required to determine those attributes within a reasonable time.
There have been numerous algorithms designed for attribute reduction from the decision table. The most typical approaches use rough set (Xia et al., 2020) (Abdolrazzagh-Nezhad et al., 2020), fuzzy rough set (Ding et al., 2020) (Yuan et al., 2021), fuzzy set (Hu et al., 2021), etc. Intuitionistic fuzzy set (IFS) (Atanassov, 1986) is considered a general model of the fuzzy set with the membership and non-membership functions to calculate the similarity and difference objects in the target set. Some metrics were generated on IFS, such as the intuitionistic fuzzy dependence (Tiwari et al., 2018 & 2019) (Tan et al., 2019) (Singh et al., 2019), intuitionistic fuzzy conditional entropy (Tan et al., 2020), granular computing (Tan et al., 2020). From those metrics, the Filter or the Filter-Wrapper approaches have been designed to find the reducts. It has been known that data may contain outliers that affect outcomes of the attribute reduction process. IFS has been regarded as an effective tool to deal with such a problem by adding two degrees: the membership and non-membership for each data element. The separation of attributes in the view of two counterparts as in the IFS set would increase classification quality and reduce the reducts. Figure 1 shows a picture of the related approaches for attribute reduction.
Figure 1. Classification of the related approaches for attribute reduction
This research is motivated by the following drawbacks of the IFS approach as follows.