1. Significance of the Study
A decision support system is a computerized system used to help make decisions, judgments and courses of action in companies or organizations [
1]. A decision support system (DSS) searches and analyzes large amounts of data and compiles extensive data for problemsolving and decisionmaking. The standard DSS information includes the actual or estimated profits, revenue and past statistics from different periods, and other stock or administrative details [
2]. A network connected with the transmission control protocol/internet protocol (TCP/IP) to the computer server runs a DSS request. The decision support system can be a communication, datadriven, knowledgedriven, modeldriven or hybriddriven Webbased DSS [
3]. Any class or type of decision support system may use web technologies. The Webbased DSS means the whole software is deployed with Web technologies; Webenabled means that the most important parts of a framework such as a server stay in an old system, but a Webbased portion can provide access and display on a Webbased interface [
4].
Figure 1 shows the flow chart of the decision support system based ecommerce model. The main aim of using a DSS is to provide the customer with easily understood knowledge [
5]. A DSS is useful because many different types of reports can be configured, all based on user preferences [
6]. As in the case of a bar chart that is reflecting projected revenue or a written report, the DSS can graphically generate data and generate its information. Data analyzes are not limited to huge, large mainframe computers anymore as technology continues to progress. As the DSS is essentially an application, it can be indicated either on desktops or laptops on most computer systems. Mobile devices offer certain DSS applications [
7,
8,
9].
A DSS can be adapted for all industries, professions or domains, including medical, government agencies, farms, and businesses [
10,
11]. The DSS can be used by managing operations and other planning divisions of an enterprise to gather and synthesize data and information efficiently. The key use of these systems is mid to top management. In electronic commerce, a decision support system gathers and analyzes data to create comprehensive information about customers. It varies from a normal operations application, the purpose of which is only to collect data, as an information application [
12,
13].
The distinction between conventional businesses and ecommerce businesses is that the ecommerce approach integrates information technology (IT) and communications and business processes, making it easier for consumers to do business. The process of buying and selling goods and services over the internet is usually electronic commerce. It is one way of sharing the information between individuals and organizations by integrating a variety of processes, including electronic data exchange (EDI), electronic mail, the World Wide Web, and electronic funds transfers (EFT) [
14,
15]. In the case of appropriate concepts, software techniques and languages, fuzzy logic is seen as a management approach that creates an effective platform for analysis and business regulation [
16]. The main contribution of fuzzy to machine learning is its incremental capacity to present gradual concepts and features. Fuzzy definitions in a specific problem domain are used as modelling components [
17]. When a problem domain is found in machine learning, it can be broken down gradually.
This paper proposes an operational researchbased intelligent decision support system (ORIDSS) for ecommerce decisionmaking using fuzzy logic theory. The decision supports a system in ecommerce that allows for scheduling and transportation optimization, data collection, improving market operations, conducting a risk analysis, matching buyer to sellers, and assisting in running Business to Consumer (B2C) operations. An important feature called the automatic product recommendation system is required for the online selection of products. As social networks on the internet are becoming more common, users cannot get detailed product or service information using their previous customers’ views. The decisionmaking process has been influenced by reliable people’s data and not by consumers’ distributors or recommendations.
The main contribution of the paper is:
Designing the ORIDSS model for Ecommerce utilizing the fuzzy logic theory and machine learning methods.
Evaluating the mathematical model for decision making in Ecommerce.
The experimental results have been executed, and the proposed model enhances the performance, accuracy, precision and reduces the error rate compared to other existing models.
The article is organized as follows:
Section 1 and
Section 2 explain the existing methods and theoretical study. In
Section 3, the mathematical model of an intelligent decision support system has been demonstrated. In
Section 4, the experimental results have been discussed. Finally,
Section 5 concludes the research paper.
2. Literature Review
Hong Zhou et al. [
18] proposed the big databased intelligent decision support system (BIDSS) for sustainable regional development. The system is appropriate for advanced planning, cooperation, and management by government agencies and companies. This incorporates stateoftheart multidisciplinary technologies such as data mining, decisionmaking artificial intelligence and communications. The system uses large amounts of data from various sources such as nonprofit organizations, governments, and companies in numerous forms such as multimedia and text. One possible algorithm to help people in various positions make decisions based on others’ behavior is the generalized Kuhn–Tucker approach to bilevel programing.
ChienChih Yu et al. [
19] introduced the Webbased consumeroriented intelligent decision support system (CIDSS) for personalized eservices. It facilitates all stages of customer decisionmaking in eservices applications between businesses and consumers. The system framework’s key functional modules include customer and customized management, navigation and browsing, assessments, development, community management, cooperative management, auctioning and payments, transactions and negotiation, performance and input control, and communications distribution of data. Therefore, it can finally lead to customer relationships and bring values and resources into the entire value chain by providing customers with great satisfaction.
Osama sohaib et al. [
20] initialized the multicriteria group decision making (MCGDM) for the Ecommerce enterprise decisionmaking system. They used 2tuple fuzzy linguistic decision making to address the multicriteria decisionmaking problem. The MCGDM is a combination of the MCDM and group decisionmaking methods that efficiently make a final decision in a group. To define several appropriate requirements, the model proposed is based on a technologyorganizationenvironment (TOE). A small to mediumsized firm uses the approach suggested to promote the evaluation of cloud ecommerce variables and make decisions.
ChangShing Lee et al. [
21] initialized the meeting scheduling decision support system using an intelligent fuzzy agent (IFA). For the final meeting time, the MSDSS collects the meeting details and sends the meeting host’s appropriate time. An intelligent fuzzy agent (IFA) is suggested to carry out an effective meeting schedule selection with a meeting scheduling agency (MNA), a fuzzy inferential agent (FIA) and a genetical learner agent (GLA). The meeting negotiation agent sends information on the meeting, including the meeting’s length, the invitee’s names and priorities, meeting the event’s importance, and meeting the time requirements with FIA working priorities.
Leung et al. [
22] introduced the fuzzy association rule mining approach (FARM) for a pricing decision support system. The business environment of ecommerce mainly reshapes the actions of B2C customers purchasing, not B2B. Although B2B consumers’ origin and the purchasing process can all be done through B2B ecommerce sites, B2B customers who make an online application often activate the RFQ process. With the growing number of factors that can be considered in today’s B2B ecommerce world, the difficulty in price determination increases.
Pinter et al. [
23] suggested the call detail records and hybrid machine learning approach (CDRHMLA) for modeling real estate prices. The prediction model is based on a multilayered perceptron (MLP) machine learning system trained with the partial swarm’s evolutionary optimization algorithm (PSO). The model’s efficiency is assessed employing an MSE, sustainability index and Willmott’s index (WI).
In this paper, an operational researchbased intelligent decision support system (ORIDSS) for ecommerce decisionmaking using fuzzy logic theory has been proposed to overcome these issues. In this paper, general guidelines and rules governing the development of key factors affecting the improvement of emotionally supportive networks and selection models should be analyzed, prepared and reviewed. Ecommerce systems are large systems that produce a great deal of data collection, especially concerning customers’ behavior. For management and decision support, information may have a high value. Using appropriate tools, the information must be acknowledged, analyzed and properly displayed.
3. Proposed System (Operational ResearchBased Intelligent Decision Support System)
In this paper, an operational researchbased intelligent decision support system (ORIDSS) for ecommerce decision making using fuzzy logic theory has been proposed. Fuzzy set theory has become a more common approach to reflect, interact with, and successfully employed in many information technology contexts, including control technology, intelligent decision support system (IDSS) and soft computing. Ecommerce is characterized as an attempt to increase transactional efficiency and effectiveness by using current and emerging digital technology in all aspects of the production, design, sales, and marketing of products or services for established and developing markets. In globalization, understanding the adoption by developing ICT countries involving ecommerce is becoming crucial to enhance its adoption. This, in turn, enables more efficient trade between the developed countries and developing countries.
Figure 2 shows the basic ecommerce system. The ecommerce system can be described as a web server linked to the information system of the business. The clear concept of the ecommerce system derives from the description of an information system whose fundamental factors are communication systems and information and users. Information systems are specifically designed to support management.
This research suggests a fuzzy logic based systemic and learning approach to promote price negotiation in ecommerce organizations. This study facilitates the understanding of the onetoone price problem in the bilateral negotiation. The model is designed to provide the right customer with the right price. Assume that
$Y=\left({y}_{1},\dots {y}_{m}\right)$ and the price offered is
x, and the problem given is to build the mathematical link between price
x and the influencing vector of
Y. It is therefore referred to as:
Preposition 1: If the Q (Y) relation can be learned from historical information, then the correct bargaining price can be calculated for each new customer using Q (Y) based on its influential factors’ values. As uncertainties are inherent in the pricing of contracts, they may occur concerning influence or the relationship between the factors of influence and the value proposed. A fuzzy logic system solution is, therefore, acceptable and useful.
The mathematical relationship between the m input variable and one output variable. A complete fuzzy rule base of the standard fuzzy logic system needs
$L={{\displaystyle \prod}}_{i=1}^{m}{M}_{i}$ fuzzy rules, where
${M}_{i}$ is the number of fuzzy sets of the ith input variable. Given input
$Y=\left({y}_{1},\dots {y}_{m}\right)$ and output
$\widehat{x}$. The standard fuzzy logic system can be illustrated as the following Equation (2),
As shown in Equation (2), where J is the index set that defined as $J=\{{j}_{1},\dots {j}_{m}{j}_{i}=1,2,..{M}_{i};i=1,2,..m\},{j}_{1}{j}_{2},\dots {j}_{m}$ is the index of fuzzy rules and ${\mu}_{{j}_{i}}^{i}\left({y}_{i}\right)$ membership degree of the jth fuzzy set in the ith input variable.
Preposition 2: Single input and single output standard fuzzy system expressed for the output and various important weight that can be expressed as the following Equation (3),
The hierarchical fuzzy system in this paper represented as the following Equation (3),
The hierarchical fuzzy system contains several standard fuzzy systems. As shown in Equation (4) where
${x}_{1}^{k,a},\dots .{x}_{{m}_{k,a}}^{k,a}$ denotes the intermediate attributes, y denotes the original input variable, while x denotes the output of the sub fuzzy system.
Figure 3
shows the modified structure of the fuzzy system.
By employing the triangular membership function in the hierarchical fuzzy system can be derived as in Equation (5),
Equation (4) can be remodified as in Equation (6),
The hierarchical fuzzy system can be shown as the number of fuzzy rules,
As shown in Equation (7), where P is the number of points in the hierarchy fuzzy system and ${w}_{j}$ is the number of sub fuzzy system in the jth level, ${m}_{j,i}$, and ${n}_{j,i}$ correspondingly.
Preposition 3: A gradient descent learning algorithm has been used in the fuzzy logic system. The gradient descent algorithm’s objective function only reduces the error based on the current information example,
This algorithm aims to reduce the local error by updating the attributes and implements a standardization step for handling the intermediate variable in the hierarchical fuzzy system. Because of a hierarchical fuzzy system, the final error back propagates from the edge to the lower levels, and the upperlevel error is used to change the corresponding lowlevel parameters. Assume r is the index of the iteration of learning; the final error of a hierarchical fuzzy system is as follows in Equation (9):
As shown in Equation (9) Where
${o}^{P,1}\left(r\right)$ is the expected output of the toplevel sub fuzzy system in the the training iteration, P is the total number of levels in the hierarchical fuzzy system, and while x(t) is the target output of the data set. Provided the structure of the hierarchy. The error is propagated to sub fuzzy system as follows:
The k+ 1 and k is omitted for the level index the Equation (10) remodified as in Equation (11),
Therefore, the hierarchical fuzzy system can be simplified as the following Equation (12),
The output of the sub fuzzy system is illustrated as the following Equation (13),
Equation (11) can be represented as the following Equation (14),
Equation (14) can be remodified as the following Equation (15),
It is derived from the concept of triangular membership functions and expressed as the following Equation (16),
As shown in Equation (16) where ${\mathrm{s}}^{\mathrm{il}}$ denotes the central point of the ith fuzzy set of the lth attribute in sub fuzzy system.
This makes it possible to reflect the propagated error as,
Because of the previous iteration outcomes, the
${x}_{{J}_{b}}^{b}\left(r\right)$ gradient descent learning algorithm updates the current iteration parameters by iteratively,
As shown in Equation (18), where
$\mathsf{\lambda}$ is the parameter of the learning rate.
Equation (18) can be remodified as the following Equation (20),
The proposed operational researchbased intelligent decision support system for ecommerce decision making using a fuzzy logic system achieves minimum error rate while deciding for the customer. The consumer conceptual model helps to reduce the difficulty level and makes it efficient.
In this paper, the researchers propose a fuzzy logic system on the operational research based on an intelligent decision support system for ecommerce decision making. The gradient descent learning algorithm has been proposed for the hierarchical fuzzy system input and output for the pricing negotiation in ecommerce. Gradient descent is an optimization algorithm used in different machine learning algorithms to minimize the cost function (List on Algorithm 1). It is mainly used to modify the learning model’s parameters. This is a method of downward gradient processing 1 example of training per iteration. Therefore, even after one iteration in which only one instance has been processed, the parameters are modified. This is much slower than the descent of the batch gradient. When the number of training examples is high, only one example is processed, which can be additional overhead for the system as the number of iterations will be quite large.
Algorithm 1: Gradient Descent Learning Algorithm 
Input: i,j, l,k, h

Output: ${x}_{{\mathrm{J}}_{a}}^{a}$, ${o}^{a}\left(r\right)$ 
For (i = 0) 
$\widehat{x}=f\left(Y\right)={\displaystyle {\displaystyle \sum}_{{j}_{1,\dots {j}_{m}\in J}}}\left({\displaystyle {\displaystyle \prod}_{i=1}^{m}}{\mu}_{{j}_{i}}^{i}\left({y}_{i}\right)\right){x}_{{j}_{1}},\dots {j}_{m},$ 
For (j = 0) 
${o}^{k,a}={\displaystyle {\displaystyle \sum}_{{i}_{1}{i}_{2}\dots {i}_{{m}_{k,a}}{j}_{1}{j}_{2}\dots {j}_{{n}_{k,a}}}}\left[{\displaystyle {\displaystyle \prod}_{l=1}^{{m}_{k,a}}}{\mu}_{il}^{k,a,l}\left({x}_{l}^{k,a}\right){\displaystyle {\displaystyle \prod}_{l=1}^{{n}_{k,a}}}{E}_{{j}_{l}}^{k,a,l}\left({y}_{l}^{k,a}\right)\right]{x}_{{i}_{1}{i}_{2}..{i}_{{m}_{k,a}}{j}_{1}{j}_{2}\dots {j}_{{n}_{k,a}}}^{k,a}$ 
For (l = 0) 
${h}_{a}\left(r\right)={h}_{b}\left(r\right)\times \frac{\partial {o}^{b}\left(r\right)}{\partial {o}^{a}\left(r\right)}$ 
If (h = 0) 
$G={\displaystyle \sum}_{j=1}^{P}\left[{\displaystyle {\displaystyle \sum}_{i=1}^{{w}_{j}}}\left({\displaystyle {\displaystyle \prod}_{l=1}^{{m}_{j,i}}}{M}_{j,i}^{l}{\displaystyle {\displaystyle \prod}_{l=1}^{{n}_{j,i}}}{M}_{j,i}^{l}\right)\right]$ 
Else 
${o}_{{\mathrm{i}}_{1}{i}_{2}}^{a}=\left[{\displaystyle {\displaystyle \prod}_{\mathrm{l}=1}^{{m}_{a}}}{\mathsf{\mu}}_{{\mathrm{i}}_{\mathrm{l}}}^{a,\mathrm{l}}\left({x}_{\mathrm{l}}^{a}\right){\displaystyle {\displaystyle \prod}_{k=1}^{{n}_{a}}}{E}_{{j}_{\mathrm{l}}}^{a,\mathrm{l}}\left({y}_{\mathrm{l}}^{a}\right)\right]{\mathrm{x}}_{{\mathrm{i}}_{1}{\mathrm{i}}_{2}\dots {\mathrm{i}}_{{\mathrm{m}}_{\mathrm{a}}}{\mathrm{j}}_{1}{\mathrm{j}}_{2}\dots {\mathrm{j}}_{{\mathrm{n}}_{\mathrm{a}}}}^{a}$ 
End for 
End for 
End for 
End if 
End 
Return 