Multi-source information fusion based on rough set theory: A review

Highlights

• Homogeneous and heterogeneous MSIF models are systematically introduced.

• Multi-view rough set is discussed by combining several rough set models.

• Parallel computing model based rough sets is reviewed in view of information fusion.

• Incremental learning fusion technology based rough sets is surveyed.

• Cluster ensembles and three-way clustering fusion approaches based rough sets are overviewed.

Abstract

Multi-Source Information Fusion (MSIF) is a comprehensive and interdisciplinary subject, and is referred to as, multi-sensor information fusion which was originated in the 1970s. Nowadays, the types and updates of data are becoming more multifarious and frequent, which bring new challenges for information fusion to deal with the multi-source data. Consequently, the construction of MSIF models suitable for different scenarios and the application of different fusion technologies are the core problems that need to be solved urgently. Rough set theory (RST) provides a computing paradigm for uncertain data modeling and reasoning, especially for classification issues with noisy, inaccurate or incomplete data. Furthermore, due to the rapid development of MSIF in recent years, the methodologies of learning under RST are becoming increasingly mature and systematic, unveiling a framework which has not been mentioned in the literature. In order to better clarify the approaches and application of MSIF in RST research community, this paper reviews the existing models and technologies from the perspectives of MSIF model (i.e., homogeneous and heterogeneous MSIF model), multi-view rough sets information fusion model (i.e., multi-granulation, multi-scale and multi-view decisions information fusion models), parallel computing information fusion model, incremental learning fusion technology and cluster ensembles fusion technology. Finally, RST based MSIF related research directions and challenges are also covered and discussed. By providing state-of-the-art understanding in specialized literature, this survey will directly help researchers understand the research developments of MSIF under RST.

Introduction

The methodology of Multi-Source Information Fusion (MSIF) is based on multi-sensor information fusion, which can achieve more significant information with higher accuracy than a single source (or a single sensor). In the late 1970s, the word of fusion based on the comprehensive meaning of multi-source information began to appear in various publications. Since then, the theories and technologies of MSIF have developed rapidly, as an independent discipline, which has been successfully applied in military command automation system, strategic warning and defense system, multi-target tracking and identification. Moreover, MSIF is also gradually radiating to remote sensing monitoring, medical diagnosis, electronic commerce, wireless communication and fault diagnosis and other civilian fields [1], [2], [3], [4].

The amount of data produced in the world each year is rising at a rate of thirty percent [5]. Data is produced by everything around us via social media exchange, and transmitted by all kinds of networks, sensors, and mobile devices. Data acquiring is no longer limited to a single data source with the full apperceive of information in a Big Data environment. The storage and description of data appears in the form of multiple sources. Various information of knowledge structure are implied in the relationships among data samples from different data sources, which express information among data samples from multiple perspectives. Fortunately, the fundamental principle of MSIF is to make full use of multiple information sources. According to specific criteria, we can combine multiple sources of information with constraints of spatial redundancy, temporal redundancy or complementary information. MSIF is studied widely in different real-life applications and diverse theories and methods have been used in MSIF. For instance, Zhang et al. proposed a multiple-metric learning algorithm to learn jointly a set of optimal homogeneous or heterogeneous metrics in order to fuse the data collected from multiple sensors for joint classification [6]. Dasarathy described a panoramic overview of MSIF in the field of multi-sensor from three complementary perspectives [7]. Yang et al. presented a mixed structure multi-mode data fusion based on D–S evidence theory and subjective Bayesian algorithm [8]. Cai et al. proposed a data fusion method for fault detection by utilizing Bayes network model, and applied the method to the parallel simulation to improve the diagnostic accuracy of ground source heat pump system [9]. Gravina et al. presented the motivations and advantages of multi-sensor data fusion and parameters affecting data fusion, and further discussed the design choices of data fusion affected by parameters at different levels [10]. Yager introduced a general framework for the multi-source data fusion processing and algorithm development [11], and investigated a novel monotonic set measure as a means of representing the multi-source fusion imperative [12]. Li et al. addressed a method of multi-source data clustering on account of homogeneous observations which are applied for multi-target detection from cluttered background with misdetection [13]. Saadi et al. presented a framework to allow intelligent merging of multiple data sources and can be applied to the urban transportation [14].

Rough Set Theory (RST), put forward by Pawlak [15], [16], provides an effective mathematical tool for dealing with uncertainty. RST plays a critical role in extracting useful features, simplifying information processing, studying expression learning, and finding imprecise and uncertain information. At present, RST has been successfully applied to machine learning, decision analysis, process control, approximate reasoning, pattern recognition, data mining and other intelligent information processing fields [17], [18], [19], [20], [21]. From the perspective of data analysis, the main advantages of RST can be summarized as five points [15]. $\left(i\right)$ It does not require any prior knowledge of the data. $\left(ii\right)$ It has a certain ability to search the smallest collection of the data. $\left(iii\right)$ It can evaluate the significance of the data. $\left(iv\right)$ It allows the use of both qualitative and quantitative data, and $\left(v\right)$ it can generate the set of decision rules from the data.

Due to the superiority of RST, increasing effort has been directed to the study of RST based data analysis, especially the field of data fusion. The fusion process of MSIF refers to deal with the data based on different models and approaches, which is essentially information fusion. The interest in the field of MSIF based on RST has significantly increased, leading to a growing number of techniques and methods. For example, Khan and Banerjee proposed a concept of multiple-source approximation systems based on Pawlak approximation spaces, which is a precursor for researchers to study MSIF in RST [22]. Li and Fei discussed a method of information fusion in wireless sensor network [23]. Liu et al. addressed a framework for performance test and evaluation of the multi-sensor data fusion with $C^{3}I$ applications background [24]. Li et al. proposed a weighted fusion approach based on Granular Computing (GrC) and RST, which has been applied to road safety indicator analysis [25]. Yao et al. came up with a multi-source alert data understanding scheme for security semantic discovery [26]. In order to solve the fusion problem of multi-source information system (MsIS), Xu et al. presented the internal-confidence and external-confidence degrees to estimate the reliability of each information source [27], then they also considered the information fusion issue based on information entropy in fuzzy incomplete information systems [28]. Che et al. employed three approaches to address the information fusion and numerical characterization of the uncertain data [29]. Yang et al. developed a multi-granulation method for information fusion [30]. Sang et al. discussed the three kinds of multi-source decision methods based on the uncertainty of decision-making process [31]. In addition, Huang et al. addressed a new fusion method based on fuzzy information granulation, which can translate multi-source interval-valued data into trapezoidal fuzzy granules [32].

In literature, a detailed information fusion survey paper on RST [33] was published in 2019, which gives a general overview of the state-of-the-art approaches. It provides a hybrid-view about information fusion from five primary perspectives, objects, attributes, rough approximations, attribute reduction and decision making. The survey paper is comprehensive and can be a good introduction to information fusion. However, information fusion is essentially a process of integrating multi-source information from multilevel and multifaceted. In recent years, the number of proposals in the area of MSIF have significantly increased. There is therefore a gap in the current literature that requires a fuller picture of established on MSIF models and technologies. It is essential to review the past research focuses and give the most recent research trends about MSIF. Hence, this paper aims to present a comprehensive survey of the five major aspects of MSIF base on RST: MSIF fusion models (including homogeneous and heterogeneous MSIF model), multi-view rough sets information fusion model, parallel computing information fusion model, incremental learning and cluster ensembles fusion technologies, as shown in Fig. 1, and a discussion about the new trend of MSIF research.

The main contributions of this review can be summarized as follows.

$\left(1\right)$ It perceptively summarizes MSIF research achievements and clusters the research into five categories: MSIF fusion models, multi-view rough sets information fusion model, parallel computing information fusion model, incremental learning and clustering ensembles fusion technologies (Fig. 1);

$\left(2\right)$ It considers the MSIF models according to two perspectives: homogeneous and heterogeneous models;

$\left(3\right)$ It combines several rough set models to study MSIF from different perspectives, and collectively refers to these models as a multi-view rough set model;

$\left(4\right)$ It introduces the parallel computing model, which can accelerate the data fusion and be good at dealing with large-scale data via MapReduce framework;

$\left(6\right)$ It uncovers an incremental learning fusion technique, such as Incremental Learning Information Fusion (ILIF) under new immigrating multiple objects, attributes and attributes values. Moreover, it identifies related research involving incremental learning;

$\left(7\right)$ It reviews cluster ensembles fusion technologies based RST, as well as the methods of roughness and rough k-means for clustering. In addition, it emphasizes the concepts and applications of three-way clustering;

$\left(8\right)$ It suggests several emerging research topics and potential research directions in this area.