Deep Learning Algorithms and Multicriteria Decision-Making Used in Big Data: A Systematic Literature Review

,


Introduction
With the increase in population, communication of different devices in networks, Internet of ings, sensors, actuators, and so on, the size of data is increasing. is increase goes into different shapes such as volume, velocity, variety, veracity, and value extracting meaningful information and insights which are challenging tasks and burning issues. According to IDC [1], in late 2011, about 1.8 ZB of data was created. Globally, per year about 1.2 ZB (10 21 ) is generated by different sources as electronic data [2]. By 2020, data are expected to 40 ZB [3]. Tools and techniques should be available to handle big data in an easy and effective way. e tools should have the capability to process the data, apply different filtering process to extract meaningful insights, cooperate, and communicate with other tools during data exchange.
Decision-making on big data based on multicriteria is one of the challenging issues for researchers and practitioners. To make decision based on appropriate numbers of choices, effectiveness, and potentiality, researchers are finding novel means to make the decision support system for the problems of various application domains of big data by using multiple criteria in integration with machine learning and artificial intelligence. Several decision support systems are available which have the potential to support decision-making activities [4][5][6][7]. Decision-making plays an important role in the success of a business or an organization. Based on multiple criteria, the decision-making is difficult as there are several available criteria for the decision to take action.
In the recent approaches for solving problems of big data, the decision support systems are facilitated with the integration of deep learning algorithms to provide a more intelligent decision support system [8][9][10]. e decision support system has several applications in diverse areas such as business [11], energy sector [12], and agriculture [13]. Various domains explain the theories and methods which come from simple to more advanced and intelligent models for decision making [14,15].
Keeping in view the effectiveness and importance of adopting the decision support system (DSS) along with the use of deep learning algorithms, the proposed study presents a systematic literature review of the last five (05) years from 2016 to 2020. e aim of the proposed study is to identify the trends, scope, and methods from the existing literature in order to present an improved decision support system with integration of deep learning solutions to tackle big data. Keywords were defined based on the research questions for conducting the survey, and the papers were included based on the criteria of  ("Deep learning" OR "Machine learning") AND ("multi-criteria" OR "MCDM") AND ("big data" OR "Data")   Experimental results of the proposed study show that deep learning algorithms and decision support systems are exploited in a useful way. e research integrated power machine learning as well as artificial intelligence and multicriteria decision-making models to provide effective solutions to problems which are more complex. e organization of the paper is as follows: next section represents the research method with the details of the protocol followed for conducting the proposed research. Section 3 represents the results and discussion section of the paper with the analysis of answering the research questions. e paper concludes in Section 4.

Research Method
Systematic literature review is a protocol of conducting literature review of a specific research and systematically collects the research papers based on the defined research questions and keywords, analyzing the selected studies in order to assess the applications and impact of the study.   No. of paper

Complexity
Several guidelines are suggested by the researcher for conducting a systematic literature review. e guidelines are followed in the proposed system literature survey [16,17]. e following sections briefly show the research method and protocol followed for conducting the proposed study.

Search Process.
e search process is very tricky due to the reason that a paper should not be missed. Six popular libraries that are ACM, IEEE, ScienceDirect, Springer, Taylor and Francis, and Wiley Online were considered for the search process. Initially, research questions were defined. e reason behind the selection of these libraries is that these libraries are dominant in the field and publishing quality research work which are mostly peer reviewed. After that, the following keywords were defined with different Boolean operators to search the query: ("Deep learning" OR "Machine learning") AND ("multi-criteria" OR "MCDM") AND ("big data" OR "Data") Searching the libraries by the individual keywords is not an effective way to select appropriate materials. So, for this, the keywords were combined by logical  operators such as "AND" and "OR" to formulate effective queries. e formulated query for search is then applied on each library to select papers that are published in the last 5 years (2016-2020). As only single keyword is not enough to identify relevant research studies, similar keywords are identified for each of the keyword term.

Study Selection.
e designed query was searched in the specified libraries in order to obtain the relevant associated materials for the proposed study. Among these libraries, the filtering process was performed for the year range from 2016 to 2020. Using this procedure, query is applied and search results are obtained, which is shown in detail in Table 1.
From each of the libraries, the papers were filtered by title, abstract, and content. All the papers are then analysed based on its contents, and the most final set of relevant papers is selected. e final selected papers are analysed to answer the research questions and to achieve the objectives of the study. Figure 3 shows the search and filtering results in the given libraries. Figure 4 shows the initial results of the journal/magazine names along with the total number of publications in the ACM library. Figure 5 shows the proceeding names with the number of publications. Figure 6 shows the content types along with the number of publications. Figure 7 shows the publication topics along with the number of papers in the IEEE library. Figure 8 shows the number of publications in the given year in the ScienceDirect library. Figure 9 shows the article type along with the number of papers. Figure 10 shows the publication title with the number of papers. Figure 11 shows the publication disciplines in the Springer library. Figure 12 shows the publication type with their number of papers. Figure 13 shows the number of selected papers in all the given libraries including ACM, IEEE, ScienceDirect, Springer, Tailor and Francis, and Wiley Online Library. e final list of the selected papers is given in Figure 14.

Quality
Assessment. e quality of the selected papers was aimed to check for showing the relevancy of the  8 Complexity selected papers to the proposed research. Scores of 0, 0.5, and 1 were adopted to give a score to the research questions based on the specific research paper. e score 0 was given to the paper which do not satisfy the research question, 0.5 for the paper which partially satisfies the research question, and 1 for paper which fully satisfies the research questions. e quality evaluation of selected papers is depicted in Figure 15.

Results and Discussion
Big data contains a huge amount of information which needs to be extracted for a specific purpose of the user.
Such kind of data is very difficult to manage, organize, and structure. Various tools are used for extracting meaningful information and insights. e applications of machine learning algorithms play an important role in real life. e machine learning has been used mostly for classification purposes [18]. Figure 16 shows the 5 Vs of big data.
Several types of analysis were done to show the impact, number of publications, and the increase in research activities in the popular libraries. In this analysis, maybe some relevant papers are missed due to the search query and some other reasons. Figure 17 shows the number of publications in the given libraries.  Complexity Figure 18 shows the type of publications along with the number of papers in the given libraries. Figure 19 shows the year and type of publication. Figure 20 shows the number of papers in the given year. Table 2 shows the details of the answers to the research questions with the method used along with the description.    [19] Support vector machine for classifications of biomedical signal on the platform of IoT e authors proposed an approach for observing the signal using the digital signal processor and then measured the heart rate, blood oxygen saturation, and blood pressure. SVM is used for classification purpose, and it showed the data into unhealthy, healthy, and very unhealthy and described the accuracy of prediction classification.
√ √ [20] Treatments of hypertension based on big data using machine learning (ML) Application of ML algorithms for big data in achieving insights into the management of hypertension disease. e decision tree and neural networks are used as ML techniques to identify the factors to contribute in hypertension drug treatment.   [31] Healthcare big data security through machine learning e study proposed an integrated technique based on masking encryption, granular access control, activity monitoring, dynamic data encryption, and models of endpoint validation. e proposed system results by providing an efficient diagnostic system for disease in the healthcare-based big data system. Methodology of design science research is used. Literature review and application of comparison of the software architecture analysis method and existing big data reference architecture are found and compared. e AHP was used, and the experiment was done through the real-world use case.
√ √ [41] Big data and visual analytics Analytics of big data is used for solving the damages caused with alarms. e method integrates technique of interdisciplinary and accomplishing analytics with evidential inference.
√ √ [42] Video trajectory analysis using unsupervised clustering and multicriteria ranking e unsupervised trajectory method of cluster (tcluster) is proposed. e method creates object indexes by fusing high-level interpretable features, then the clusters are fused through MCDM, and trajectories are ranked accordingly.   Applying machine learning to the AHP multicriteria decision-making method to asset prioritization in the context of industrial maintenance 4.0 With the help of machine learning algorithms and relevance analysis of attribute to process the event log failure of components of industrial machine √ √ [74] Amended fused TOPSIS-VIKOR for classification e study presented a novel classification ATOVIC based on fused VIKOR and TOPSIS √ √ [75] Machine learning models, epistemic set-valued data, and generalized loss functions e research aim is to study the problem where the goal is to identify optimal model to specific criteria in supervised and regression classification problems √ √ [76] Framework for modelling of drug electrochemical removal from wastewater based on data mining algorithms, scatter interpolation method, and multicriteria decision analysis e study presented a framework for modelling of removal of drug from wastewater. With the help of data mining algorithms, ciprofloxacin electrochemical removal modelling is done and MCDA is developed for ranking the algorithm of data mining.
√ √ [77] Feature selection based on multicriteria on costsensitive data with missing values An evaluation system based on multicriteria is proposed for evaluating features from diverse viewpoints √ √

Conclusion
With the current advancements in technology, the data are ever increasing with the increase in population, communications, IoT, actuators, sensors, and so on. is increase goes into different shapes of data whose decision becomes a challenging issue due to its growth of volume. Deep learning algorithms and multicriteria-based decision-making have effective applications in big data. Due to its effectiveness and potentiality, it is exploited in several domains such as computer science and information technology, agriculture, and business sector. e aim of the proposed study is to present a systematic literature in order to show the applications of deep learning algorithms and multicriteria decision approaches for the problems of big data. e research finds novel means to make the decision support system for the problems of big data using multiple criteria in integration with machine learning and artificial intelligence approaches. e presented study will provide better insights into the research community in the area domain and will help them in the designing of more robust, efficient, and effective multicriteria-based decision support system models, framework, technique, and integrated solutions of machine learning algorithms.

Data Availability
No data are available.

Conflicts of Interest
No conflicts of interest exist regarding this paper.  [78] Frameworks of ML and data mining for prediction of drug response in cancer e study gives an overview of the supervised and unsupervised algorithms used in the prediction of drug response, the strategies applied for designing these algorithms into functional models, data resources for feeding to frameworks, and challenges for maximizing the performance of the models √ √ [79] Data fusion and machine learning for industrial prognosis e study overviewed the literature in data fusion and analysis for industrial prognosis √ √ [80] A supervised machine learning approach to datadriven simulation With the help of data analytics, the risk profiles of supplier performance based on uncertainty are analysed √ √ [81] Machine learning model-based favourite data to analyze asymmetric competition e study presented an approach for analysis of asymmetric competition based on favourite data with the help of machine learning algorithms √ √ [82] Machine learning powered software for accurate prediction of biogas production e study presented machine learning approaches for biogas production data from projects of China √ √ Complexity 15