Incremental Learning Researches on Rough Set Theory: Status and Future

Introduction

As an effective mathematical tool, rough set theory (RST) proposed by Pawlak (1982), plays an important role in many fields of data analysis. In RST, it utilizes the lower and upper approximations to deal with an imprecise and uncertainty information (concept), and has been successfully applied in many domains, such as water demand prediction (An et al., 1996), business failure prediction (Dimitras et al., 1999), cluster analysis (Lingras et al., 2010), airline service strategies (Liou et al., 2010), etc. Based on the two approximations, the main researches of RST focus on three aspects, the algebraic structure with two approximation operators (Yao, 1996, 1998); attribute reduction (feature selection) (Dey et al., 2011; Liang et al., 2013; Wang et al., 2013a; Wang et al., 2013b; Xu et al., 2011) and decision rule induction (Blaszczynski & Slowinski, 2003; Fan et al., 2009; Greco et al., 2004; Liang, 2010; Shan & Ziarko, 1992).

Unfortunately, the existing researches on rough sets mainly involve a static environment. In the real life, we may face to a dynamic decision environment (Michalski, 1985). For example, a company always encounters some employees leave or recruits this company. In order to keep the pace of produce process, the employer (manager) need decide how to employ in real time. In addition, accompanied by cheaper storage, evolution of digital data and information collection devices, such as cell phones, laptops, and sensors, the data in today’s is getting increasingly larger and need real-time processing (Demirkan & Delen, 2013). With an overwhelming amount of data arriving at a terabyte and even exabyte scale, the big data (or the massive data) becomes one of the biggest challenges in knowledge discovery. In the context of big data in dynamic environments, the existing information processing technologies based on rough sets should be extended to suit for the updating data (Chen et al., 2012a).