A novel hybrid procedure of PLS-SEM,ANN and fuzzy TOPSIS for online banking

Abstract

The adoption of online banking is a big challenge as well as an emergent paradigm which is evolving quickly. The study develops a new exploration model that is used to determine significant constructs of the UTAUT theory that influence the adoption of online-banking. A study was conducted through an online survey of online banking users. Three methods were used; firstly, PLS-SEM was used to determine which of the constructs have a significant impact on behavioral intention to use e-banking, secondly, an incorporated neural network model was used to classify the relative impact of significant variables obtained from PLS-SEM, and finally a hybrid procedure is incorporated from PLS-SEM to initialize the Fuzzy TOPSIS. The findings of the research paper constitute the ranked results and a comparison between them. The results of PLS-SEM and ANN analysis showed the same ranking for the constructs, while the decision making method Fuzzy TOPSIS introduced some changes in the ranking. This study presents valuable insights for the banking system to bring effective projects that increase the possibilities of using online banking.

Keywords

PLS-SEM ANN online banking fuzzy TOPSIS UTAUT

1 Introduction

The rapid expansion of the accessibility of the Internet has conduct the increasing use of online banking [1]. It has changed the banking system [2]. The latter has moved rapidly from traditional financial services towards self-service delivery channels [3]. Has appeared as one of the great electronic commerce applications [4].

Online banking is a banking channel that allows customers to accomplish financial transactions wherever the Internet is accessible anytime, anywhere [5]. It offers services such as balance enquiries, balance transfer, opening accounts, payment of bills [2] and other details connected to accounts [6].

Online banking helps the improvement of the efficiency and effectiveness of services [7] which in turn improves banks’ competitive positions [8] in advanced and dynamic banking system. There are many studies that explain the factors that influence the acceptance and use of online banking based on the perspectives of various adoption theories [9 –12]. In general, empirical studies are based on multiple regression model [13, 14], partial least square structural equation modelling (PLS-SEM) [15, 16]. These methods are conventional and guided by the approach of the influence of independent variables on dependent variables, excluding non-linear relationships between variables.

This paper proposes a multi-analytical conceptual model to predict customers’ behavior and to determine the main constructs that impact the decision to use online-banking. This model does not only include the factors of UTAUT, but also integrates two new constructs, which are CS and PS, bringing strength, significance and predictability of the results [17].

So the contribution of this study is that it projects an integrated model which incorporates the UTAUT model and the customer service and perceived security constructs.

This study addresses some challenges related with the development of an innovative and hybrid framework that combines PLS-SEM with Fuzzy TOPSIS a Fuzzy Technique for Order Preference by Similarity to Ideal Solution (F-TOPSIS) and also ANN analysis to better estimate the adoption of online banking.

The results obtained from PLS-SEM show the importance that the constructs of UTAUT have regarding behavioral intention to use the online banking. In order to have an efficient decision-making result [18, 19], has been integrated a hybrid model which uses these results as initiators of the fuzzy TOPSIS method with triangular fuzzy numbers [20]. These method belong to the group of multi criteria decision making methods (MCDM) [21, 22].

The fuzzy TOPSIS starts with the construction of the decision matrix according to the structured problem and output from PLS-SEM method. Exactly the results of PLS-SEM orient better a decision maker in the construction of this matrix. The practical model of this hybridization has been effective even for the online learning [20, 23]. From the methodological side, this paper presents a multi-method research plan integrating Partial Least Square Structural Equation Modelling, ANN analysis and Fuzzy TOPSIS.

PLS-SEM shows significant variables from the supported relationships. The significant variables are used as inputs for the neural network structure to decide the relative importance of significant constructs. The ANN technique also serves as a complementary method to the PLS-SEM method because it takes into account the non-linear connections of the model.

The outcomes of PLS-SEM guide the initializing of the Fuzzy TOPSIS procedure to rank the constructs of the proposed model in relation to behavioral intention to use online banking [20]. The use of multi-method predictive, which is based on the hybridization of 3 methods, ensures a holistic approach to online banking and brings a methodological contribution from the statistical aspect. (See the framework, Fig. 1).

Fig. 1

The framework of the research study.

2 Literature review

In information systems studies, frameworks with different constructs have been developed to explain individual’s acceptance and use of new technology, such as Theory of Reasoned Action (TRA) [24], Theory of Planned Behavior (TPB) [25], and the Technology Acceptance Model (TAM) [26], UTAUT [27]. Many studies in various contexts of technologies have evolved based on these models or have expanded them with new factors, making them more comprehensive. In the UTAUT model, all the constructs of the above models are integrated, improving the model’s explanatory power to behavioral intentions to use new technology [28]. UTAUT is robust and powerful model to explore determinants of behavioral intention to use a new technology [28].

2.1 Research model and hypotheses development

The UTAUT model will also serve as a ground to determine the factors that influence user acceptance of online banking. The UTAUT model includes the constructs: performance expectancy (PE), effort expectancy (EE), social influence (SI) and facilitating condition (FC). To improve the model’s explanatory power and for a holistic understanding of behavioral intention to use online banking, the model has been extended by incorporating the constructs customer service CS and perceived security (PS). The research model is presented in Fig. 1.

Performance expectancy denotes the degree to which a customer perceives that online banking induce greater performance, productivity in the realization of banking tasks [29]. The impact of performance expectancy on behavioral intention to use online banking has been widely provided in previous studies [17 , 30]. Based on previous literature, we suggest the following hypothesis:

H1: Performance expectancy positively affects the behavioral intention to use online banking.

Effort expectancy is the degree of ease associated with using the online banking [28]. When the use of online banking by the customer is easy, then the possibility of adoption increases [31]. Multiple studies confirm that effort expectancy have a significant effect on behavioral intention (BI) to use online banking [32 –36]. In addressing this issue is investigated effort expectancy and compared it with other constructs of model. Accordingly, the following hypothesis is proposed:

H2: Effort expectancy positively affects the behavioral intention to use online banking.

Social influence is social pressure that customers have from family members, and friends in order to use a particular technology [37]. In the online banking context, it has been found that social influence is an important construct of acceptance of online banking [29 , 37–39]. Consistent with these findings, we suggest the following hypothesis:

H3: Social influence positively affects the behavioral intention to use online banking.

Facilitating conditions are the degree to which a user considers that he has the required infrastructure to implement a new technology [40]. If users do not have the necessary tools such as computer, smartphone, internet and necessary operational skills they would have lower possibility to adopt a specific technology [41]. The studies show that facilitating condition is one of the factors that positively affects customers’ intention to use online banking [34, 42]. Consistent with prior literatures, we propose the following hypothesis:

H4: Facilitating condition positively affects the behavioral intention to use online banking.

Perceived security denotes the degree of confidence in online banking to transmit sensitive data [43]. This is the reason why perceived security is considered as essential element in the adoption of online banking [41]. Perceived security is an important factor on the intention to adopt online banking [17, 44]. In this context, we suggest the following hypothesis:

H5: Perceived security positively affects the behavioral intention to use online banking.

Customer service is the expectation that customers have that through online banking transactions will be finalized correctly, the product will be delivered on time, their emails will be answered quickly, receive personalized attention [45]. The adoption of internet banking is strongly impacted by customer service, as evidenced by various studies [45 –47]. Moreover, customer service influenced users’ perceptions of effort expectancy in utilizing internet banking, as highlighted by research conducted by [48 –50]. Extensive research performed by [50] and [34] demonstrates that customer service has a positive impact performance expectancy. Based on the above arguments, we suggest the following hypotheses:

H6: Customer service positively affects the behavioral intention to use online banking.

H7: Customer service positively affects the user performance expectancy of online banking.

H8: Customer service positively affects the user effort expectancy of online banking.

2.2 Research methodology

In this paper a hybridized methodology is employed by the integration of Partial Least Squares Structural Equation Modelling (PLS-SEM), artificial intelligence method named as Artificial Neural Network (ANN), and Fuzzy TOPSIS. The analysis is carried out in three stages. The first stage is connected with PLS-SEM which includes measurement model validation and structural model hypotheses testing.

In the second phase, the hypotheses accepted by the structural model are evaluated by ANN to show the relative impact of significant variables. In the third phase, the significant variables of the structural model according path coefficients are ranked by Fuzzy TOPSIS.

A self-administered questionnaire was established in Albania in 2022. The data accumulating procedure was conducted online through Gmail, Facebook platforms. The study was conducted with a case of 440 customers that had used online banking services of different commercial banks in Albania.

Questionnaire is prepared based on a 5-likert scale in order to collect the responses: (1) Strongly Disagree, (2) Disagree, (3) Neutral, (4) Agree and (5) Strongly Agree. PLS-SEM technique of multivariate statistics was applied to test the hypothesis of existing theories or to evolve theory. The notions of the model are unobservable and measured indirectly by multiple items which are called latent variables or constructs. The PLS-SEM method concentrates on explaining the variance of the dependent variable [51].

2.3 Measurement model

In the models of information systems, the concepts have been operationalized by a reflective measurement model [52]. The evaluation of the measurement model includes statistics: item loading, composite reliability (CR), average variance extracted (AVE), discriminant validity.

Item loading presents correlations of items with their corresponding latent variables and it is recommended that this statistic have values greater than 0.7 [53]. Composite reliability values must be greater than 0.7. An indicator that confirms convergent validity is average variance extracted (AVE). The lower limit of AVE values is 0.5 [54].

2.4 Structural model

After the statistics of the measurement model have been evaluated and all are within the acceptable limits, we continue with the analysis of the structural model. The latter involves testing the statistical significance of the hypotheses proposed for the research framework, the values of R², predictive relevance is calculated via Stone-Geisser’s Q² (predictive relevance). This indicator must have values greater than 0 for a dependent variable to present strong predictive relevance.

2.5 ANN evaluation

Artificial Neural Network (ANN) is a parallel distributed processing model which is formed of processing elements that are called neurons, and are connected between them with weighted links [55]. The ANN consists of neurons that are grouped into layers.

The most typical form of neural networks consists of input layer, hidden layers and output layer. In studies related to technology adoption, multi-layer perceptron (MLP) is usually used. The activation function is usually a non-linear function and the most common activation function that is used is sigmoid function because it has easily differentiable properties [56].

ANNs is an artificial intelligence method that is powerful, highly efficient in modelling complex relationships among input layers and output layers [57]. Based on the previous literature [58 –61], regarding technology adoption studies where neural networks are based on MLP are commonly used one or two hidden layers. The study data are divided into training set and testing set, where respectively 90% of the data are included in the training set and 10% for testing set [62].

Sensitivity analysis performance is calculated taking into consideration (ANN1, ANN2,...,ANN10) and the average of normalized importance for each ANN construct is calculated, which is called average relative importance [63]. Then the normalized importance of each independent variable is calculated as a ratio of its relative importance with the highest relative importance of all independent variables [64].

2.6 Fuzzy TOPSIS

The Technique for Order of Preference by Similarity to Ideal Solution (TOPSIS) is part of multi criteria decision making analysis methods (MCDM), which was originally developed by Ching-Lai Hwang and Yoon in 1981 [65]. The criteria weights in TOPSIS method are calculated using Ordinal Priority Approach or Analytic Hierarchy Process (AHP) [21].

An assumption of TOPSIS is to find the ideal solution via the shortest geometric distance from the positive ideal solution (PIS) and the longest distance from the negative ideal solution (NIS) [66]. Some of the criteria can be addressed as benefit criteria and cost criteria where the scope is to maximize the benefit and to minimize the cost. Using fuzzy numbers usually triangular fuzzy numbers Saaty scale [18] in TOPSIS for criteria estimation leads to Fuzzy TOPSIS method.

The optimal criteria is it that is nearest to the PIS solution. A complex decision problem is based on alternatives A₁, A₂, ... ,A_n and criteria C₁, C₂, ... ,C_m. The group of decision makers judge the construction of the decision matrix of the problem D (x_ij), where x_ij is the comparison of the alternative A_i toward criteria C_j.

Fuzzy TOPSIS steps:

Step 1: Define the fuzzy ratings of the k-th decision maker $x_{ij}^{k}$ and the weights of the criteria $ω_{j}^{k}$ .

Step 2: The evaluation of the aggregated fuzzy ratings x_ij and the aggregated fuzzy weights ω_j. $x_{ij} = (a_{ij}, b_{ij}, c_{ij})$ $a_{ij} = min_{k} a_{ij}^{k}, b_{ij} = \frac{1}{k} \sum_{k} b_{ij}^{k}, c_{ij} = max_{k} c_{ij}^{k}$ $ω_{j} = (ω_{j 1}, ω_{j 2}, ω_{j 3})$ $ω_{j 1} = min_{k} ω_{j 1}^{k}, ω_{j 2} = \frac{1}{k} \sum_{k} ω_{j 2}^{k}, ω_{j 3} = max_{k} ω_{j 3}^{k}$

Step 3: Compute the normalized fuzzy decision matrix $\overset{⌣}{R} = {{\overset{⌣}{r}}_{i j}}$ where $\begin{array}{l} {\overset{⌣}{r}}_{i j} = (\frac{a_{i j}}{c_{j}^{*}}, \frac{b_{i j}}{c_{j}^{*}}, \frac{c_{i j}}{c_{j}^{*}}) c_{j}^{*} = \max_{i} c_{i j} (b e n e f i t c r i t e r i a) \\ {\overset{⌣}{r}}_{i j} = (\frac{c_{j}^{-}}{c_{i j}}, \frac{c_{j}^{-}}{b_{i j}}, \frac{c_{j}^{-}}{a_{i j}}) c_{j}^{-} = \min_{i} a_{i j} (cost c r i t e r i a) \end{array}$

Step 4: The weighted normalized fuzzy decision matrix $V_{i j} = ({\overset{⌣}{v}}_{i j})$ where ${\overset{⌣}{v}}_{i j} = {\overset{⌣}{r}}_{i j} \cdot ω_{j}$

Step 5: Compute the fuzzy positive ideal solution (FPIS) and the fuzzy negative ideal solution (FNIS). $\begin{matrix} (F P I S) A^{*} = ({\overset{⌣}{v}}_{1}^{*}, {\overset{⌣}{v}}_{2}^{*}, \dots, {\overset{⌣}{v}}_{n}^{*}) w h e r e {\overset{⌣}{v}}_{j}^{*} = \max_{i} {\overset{⌣}{v}}_{i j}^{*} \\ (F N I S) A^{-} = ({\overset{⌣}{v}}_{1}^{-}, {\overset{⌣}{v}}_{2}^{-}, \dots, {\overset{⌣}{v}}_{n}^{-}) w h e r e {\overset{⌣}{v}}_{j}^{-} = \min_{i} {\overset{⌣}{v}}_{i j}^{-} \end{matrix}$

Step 6: Calculate the Euclidian distance from FPIS and FNIS $\begin{matrix} d_{i}^{*} = \sum_{j = 1}^{n} d_{v} ({\overset{⌣}{v}}_{i j}, {\overset{⌣}{v}}_{j}^{*}), d_{i}^{-} = \sum_{j = 1}^{n} d_{v} ({\overset{⌣}{v}}_{i j}, {\overset{⌣}{v}}_{j}^{-}), d \\ d (x, y) = \frac{1}{3} \sqrt{{(a_{1} - a_{2})}^{2} + {(b_{1} - b_{2})}^{2} + {(c_{1} - c_{2})}^{2}} \end{matrix}$

Step 7: Compute the closeness coefficient CC_i as follows: ${CC}_{i} = \frac{d_{i}^{-}}{d_{i}^{-} + d_{i}^{*}}$

3 Results

3.1 Evaluation of the measurement model

The item loadings shown in Table 1 varied from 0.759 to 0.924, so it’s confirmed that they are greater than the “minimum requirement, which is 0.7.

Table 1
Measurement model assessment

Constructs Loadings α CR AVE

and Items

(BI) 0.866 0.918 0.789

BI1 0.89

BI2 0.901

BI3 0.872

(CS) 0.841 0.894 0.678

CS1 0.759

CS2 0.815

CS3 0.856

CS4 0.86

(EE) 0.873 0.914 0.726

E1 0.855

E2 0.88

E3 0.87

E4 0.801

(FC) 0.75 0.856 0.665

FC1 0.83

FC2 0.847

FC3 0.766

(SI) 0.891 0.932 0.82

SI1 0.885

SI2 0.873

SI3 0.887

(HT) 0.86 0.905 0.705

HT1 0.876

HT2 0.875

HT3 0761

HT4 0.841

(PE) 0.853 0.9 0.693

PE1 0.818

PE2 0.849

PE3 0.835

PE4 0.828

(PS) 0.893 0.933 0.823

PS1 0.889

PS2 0.908

PS3 0.924

(UB) 0.87 0.92 0.794

UB1 0.877

UB2 0.881

UB3 0.915

Constructs	Loadings	α	CR	AVE
(BI)		0.866	0.918	0.789
BI1	0.89
BI2	0.901
BI3	0.872
(CS)		0.841	0.894	0.678
CS1	0.759
CS2	0.815
CS3	0.856
CS4	0.86
(EE)		0.873	0.914	0.726
E1	0.855
E2	0.88
E3	0.87
E4	0.801
(FC)		0.75	0.856	0.665
FC1	0.83
FC2	0.847
FC3	0.766
(SI)		0.891	0.932	0.82
SI1	0.885
SI2	0.873
SI3	0.887
(HT)		0.86	0.905	0.705
HT1	0.876
HT2	0.875
HT3	0761
HT4	0.841
(PE)		0.853	0.9	0.693
PE1	0.818
PE2	0.849
PE3	0.835
PE4	0.828
(PS)		0.893	0.933	0.823
PS1	0.889
PS2	0.908
PS3	0.924
(UB)		0.87	0.92	0.794
UB1	0.877
UB2	0.881
UB3	0.915

Moreover, the internal consistency reliability was assessed by observing that the Composite Reliability (CR) values reached 0.8, surpassing the minimum threshold of 0.7, as described in Table 1.

The values of the AVE statistic are greater than 0.5, so they reflect satisfactory convergent validity achieved by the constructs.

3.2 Evaluation of the structural model

The structural model is evaluated based on the significance of path coefficients and the variance of the dependent construct, represented by R square.

The significance of the path coefficients was evaluated through a T-test conducted using the bootstrapping technique, with a significance level set at 5%. The results of Fig. 2 and Table 2 present the path coefficients of the structural model and the testing of the hypotheses taken into consideration. Table 2 shows the path coefficients.

Fig. 2

Path coefficients of structural model.

Table 2

Path coefficients

	Path coefficients	P Values
BI → UB	0.766	0
CS → BI	0.261	0.001
CS → EE	0.713	0
CS → PE	0.626	0
EE → BI	0.152	0.001
PE → BI	0.13	0.003
SI → BI	0.147	0.004
FC → BI	0.078	0.118
PS → BI	0.205	0

The path coefficient for the Performance expectancy influence on behavioral intention to use online banking (β= 013, p = 0.05) indicated the significant positive influence of Performance expectancy on behavioral intention to use online banking. The path coefficient for influence of effort expectancy on behavioral intention to use online banking (β= 0.152, p = 0.05) revealed its significant positive impact on behavioral intention to use online banking.

The path coefficient for the influence of customer service on behavioral intention to use online banking (β= 0.261, p < 0.05) revealed its significant and positive impact.

Additionally, the path coefficient for the influence of perceived security on behavioral intention to use online banking (β= 0.205, p < 0.05) revealed its significant and positive influence, the path coefficient for the influence of social inclusion on behavioral intention to use online banking (β= 0.147, p < 0.05) revealed its significant and positive influence.

Whereas facilitating condition has no significant impact on behavioral intention to use online banking.

The path coefficients for the effects of customer service on effort expectancy (β= 0.713, p < 0.05), and on performance expectancy (β= 0.626, p < 0.05), indicated their respective significant and positive effects.

Table 3 shows that the research model achieved 66.5% of the behavioral intention variation towards online banking. The R² values are also presented for EE (50.8%), PE (39.1%), UB (60.4%).

Table 3

Variance explained from constructs

Constructs	R square
BI	0.665
EE	0.508
PE	0.391
UB	0.604

3.3 ANN analysis

In our research framework, we used SPSS v22 to model the MLP-ANN. The model that has a dependent variable behavioral intention was taken into consideration.

The significant constructs determined by PLS-SEM are used as inputs to the model.

The model has 5 important constructs. The recommendations provided by Leong [64] entail the automatic generation of hidden neurons (nodes) and the utilization of the Sigmoid function as the activation function for both the hidden and output layers.

Fig. 3

ANN model.

The results in Table 4 show that in relation to the normalized relative importance indicator, the customer service construct is the most important determinant of behavioral intention, which is followed by perceived security, effort expectancy and social inclusion.

Table 4

The normalized relative importance

Independent constructs	Average relative importance	Normalized relative, importance	Ranking
CS	90.6%	100%	1
EE	70%	66%	3
PE	49%	32.6%	5
SI	61%	56.1%	4
PS	73.6%	66.1%	2

Performance expectancy has the weakest influence on behavioral intention compared to other constructs.

3.4 Fuzzy TOPSIS

The importance of the significant constructs derived from PLS-SEM analysis (see Table 2) orients better the initialization of the fuzzy TOPSIS method in order to construct the decision matrices.

Three decision makers have evaluated the importance of each of the constructs toward the behavioral intention to use online banking. Each of the decision matrices (DM1, DM2, DM3) must be consistent according Saaty inconsistence (IC) less than 0.1.

Therefore, an integration of the two approaches PLS-SEM and TOPSIS was realized to initially evaluate the proposed model.

Table 6 shows the results for the decision matrices related to decision maker 1 (DM1), decision maker 2 (DM2), and decision maker 3 (DM3). According path coefficients the construct FC (facilitating conditions) is not taken as an important one because the p-value is greater than 0.05. Table 5 shows the initial decisions.

Table 5
Initial decisions of fuzzy TOPSIS

BI CS EE PE SI PS

CS (1, 1, 1)DM1 (6,7,8)DM1 (5,6,7)DM1 (2,3,4)DM1 (0.25,0.33,0.5)DM1

(1, 1, 1)DM2 (3,4,5)DM2 (2,3,4)DM2 (5,6,7)DM2 (0.33,05,1)DM2

(1, 1, 1)DM3 (3,4,5)DM3 (5,6,7)DM3 (4,5,6)DM3 (0.25,0.33,0.5)DM3

EE (0.12,0.14,0.16)DM1 (1, 1, 1)DM1 (1, 2, 3)DM1 (1, 2, 3)DM1 (0.2,0.25,0.33)DM1

(0.2,0.25,0.33)DM2 (1, 1, 1)DM2 (1, 2, 3)DM2 (2,3,4)DM2 (0.33,0.5, 1)DM2

(0.2,0.25,0.33)DM3 (1, 1, 1)DM3 (2,3,4)DM3 (6,7,8)DM3 (0.2,0.25,0.33)DM3

PE (0.14,0.16,0.2)DM1 (0.33,0.5, 1)DM1 (1, 1, 1)DM1 (0.33,0.5, 1)DM1 (0.16,0.2,0.25)DM1

(0.25,0.33,0.5)DM2 (0.33,0.5, 1)DM2 (1, 1, 1)DM2 (0.33,0.5, 1)DM2 (0.12,0.14,0.2)DM2

(0.14,0.16,0.2)DM3 (0.25,0.33,0.5)DM3 (1, 1, 1)DM3 (2,3,4)DM3 (0.16,0.2,0.25)DM3

SI (0.25,0.33,0.5)DM1 (0.33,0.5,1)DM1 (1,2,3)DM1 (1, 1, 1)DM1 (0.25,0.33,0.5)DM1

(0.14,0.16,0.2)DM2 (0.25,0.33,0.5)DM2 (1,2,3)DM2 (1, 1, 1)DM2 (0.16,0.2,0.25)DM2

(0.16,0.2,0.25)DM3 (0.12,0.14,0.16)DM3 (0.25,0.33,0.5) DM3 (1, 1, 1)DM3 (3,4,5)DM3

PS (2,3,4)DM1 (3,4,5)DM1 (2,3,4)DM1 (2,3,4)DM1 (1, 1, 1)DM1

(0.33,0.5,1)DM2 (1,2,3)DM2 (6,7,8)DM2 (4,5,6)DM2 (1, 1, 1)DM2

(2,3,4)DM3 (3,4,5)DM3 (4,5,6)DM3 (0.2,0.25,0.33)DM3 (1, 1, 1)DM3

BI	CS	EE	PE	SI	PS
CS	(1, 1, 1)DM1	(6,7,8)DM1	(5,6,7)DM1	(2,3,4)DM1	(0.25,0.33,0.5)DM1
	(1, 1, 1)DM2	(3,4,5)DM2	(2,3,4)DM2	(5,6,7)DM2	(0.33,05,1)DM2
	(1, 1, 1)DM3	(3,4,5)DM3	(5,6,7)DM3	(4,5,6)DM3	(0.25,0.33,0.5)DM3
EE	(0.12,0.14,0.16)DM1	(1, 1, 1)DM1	(1, 2, 3)DM1	(1, 2, 3)DM1	(0.2,0.25,0.33)DM1
	(0.2,0.25,0.33)DM2	(1, 1, 1)DM2	(1, 2, 3)DM2	(2,3,4)DM2	(0.33,0.5, 1)DM2
	(0.2,0.25,0.33)DM3	(1, 1, 1)DM3	(2,3,4)DM3	(6,7,8)DM3	(0.2,0.25,0.33)DM3
PE	(0.14,0.16,0.2)DM1	(0.33,0.5, 1)DM1	(1, 1, 1)DM1	(0.33,0.5, 1)DM1	(0.16,0.2,0.25)DM1
	(0.25,0.33,0.5)DM2	(0.33,0.5, 1)DM2	(1, 1, 1)DM2	(0.33,0.5, 1)DM2	(0.12,0.14,0.2)DM2
	(0.14,0.16,0.2)DM3	(0.25,0.33,0.5)DM3	(1, 1, 1)DM3	(2,3,4)DM3	(0.16,0.2,0.25)DM3
SI	(0.25,0.33,0.5)DM1	(0.33,0.5,1)DM1	(1,2,3)DM1	(1, 1, 1)DM1	(0.25,0.33,0.5)DM1
	(0.14,0.16,0.2)DM2	(0.25,0.33,0.5)DM2	(1,2,3)DM2	(1, 1, 1)DM2	(0.16,0.2,0.25)DM2
	(0.16,0.2,0.25)DM3	(0.12,0.14,0.16)DM3	(0.25,0.33,0.5) DM3	(1, 1, 1)DM3	(3,4,5)DM3
PS	(2,3,4)DM1	(3,4,5)DM1	(2,3,4)DM1	(2,3,4)DM1	(1, 1, 1)DM1
	(0.33,0.5,1)DM2	(1,2,3)DM2	(6,7,8)DM2	(4,5,6)DM2	(1, 1, 1)DM2
	(2,3,4)DM3	(3,4,5)DM3	(4,5,6)DM3	(0.2,0.25,0.33)DM3	(1, 1, 1)DM3

Applying equations of step 1 to step 5 is obtained the weighted normalized fuzzy decision matrix. Table 6 shows the weighted normalized fuzzy decision matrix. After calculating FPIS and FNIS is computed the closeness coefficient CC_i. Table 7 shows the ranked results.

Table 6

Weighted normalized fuzzy decision matrix

BI	CS	EE	PE	SI	PS
CS	(1.25,1.5,1.75)	(1.125,2.5,5)	(0.25,1.25,2.63)	(0.03,0.069,0.22)	(0.16,0.705,3)
EE	(0.25,0.318,0.577)	(0.38,0.5,0.63)	(0.25,0.582,1.5)	(0.027,0.083,0.44)	(0.5,1.87,3.75)
PE	(0.178,0.33,0.875)	(0.093,0.33,0.625)	(0.13,0.25,0.38)	(0.083, 0.249,1.32)	(2,3.47,6)
SI	(0.178,0.348,0.875)	(0.093,0.16,0.625)	(0.03,0.36,1.125)	(0.22,0.33,0.44)	(0.055,0.195,4.5)
PS	(0.416,3.24,7)	(0.38,1.66,3.125)	(0.5,1.414,3)	(0.03,0.123,4)	(0.5,0.625,0.75)
A ^*	(1.25,1.5,1.75)	(1.125,2.5,5)	(0.5,1.414,3)	(0.22,0.33,0.44)	(2,3.47,6)
A ^-	(0.178,0.33,0.875)	(0.093,0.16,0.625)	(0.03,0.36,1.125)	(0.027,0.083,0.44)	(0.055,0.195,4.5)

Table 7

The ranked results

	$d_{i}^{*}$	$d_{i}^{-}$	CC _i	Rank
CS	3.09	6.02	0.668	1
EE	6.91	1.79	0.205	4
PE	6.1	3.4	0.357	3
SI	7.58	0.207	0.026	5
PS	10.09	11.11	0.524	2

Table 8 shows a comparison of the results according the classification of the constructs with the ANN method and with PLS-SEM. It is showed that the constructs are ranked the same as with the PLS-SEM method and with the ANN method.

Table 8

Results comparison PLS-SEM, ANN and fuzzy TOPSIS

Constructs	Path	PLS-SEM	ANN	Fuzzy
	coefficients			TOPSIS
CS	0.261	1	1	1
EE	0.152	3	3	4
PE	0.13	5	5	3
PS	0.205	2	2	2
SI	0.147	4	4	5

4 Conclusion

This research extends the new technology adoption based on ANN that has traditionally been developed by the PLS-SEM method, as well as fuzzy TOPSIS that initiates from PLS-SEM. The impact of each predictor construct in relation to the predicted construct behavioral intention to use online banking is ranked using ANN sensitivity analysis and Fuzzy TOPSIS. The findings of the ANN paradigm confirm the results obtained from PLS-SEM, while fuzzy TOPSIS confirms the first 2 rankings while the last three change order. The combined approach of three stages PLS-SEM, ANN and Fuzzy TOPSIS method yielded more comprehensive findings concerning the relative significance of the predictor constructs. This valuable information sheds light on the adoption of the new technology. The PLS-SEM analysis shows that customer service has the strongest positive impact on behavioral intention to use online banking and ANN evaluation confirms this result by ranking customer service (CS) firstly. The integrated model showed that customer service is the most important variable among all other variables including fuzzy TOPSIS. This indicates that online banking users are more concerned about quality of customer service. Users are likely more interested in having a satisfactory customer experience, quick and helpful responses from customer support, and efficient problem resolution, which influences their overall satisfaction to the online banking platform. Perceived security was found to be the second most important construct that impacts users’ intention to adopt online banking. So customers place a high level of significance on feeling secure and safe when considering using online banking services. The differ ranks of effort expectancy, perceived expectancy and social inclusion depends from the decision matrices of the decision makers point of evaluation. The results of the study help bank managers with the necessary knowledge about the constructs of UTAUT that have an impact on behavior intention regarding the use of online banking. This study is limited because it was conducted only for the whole Albania country. The subsequent study will include in a broader geographical context and new constructs will be integrated.

References

Pouramirarsalani

, Khalilian

, Nikravanshalmani

Recognition of fraud in online banking by using confirmatory learning in neural network, Int J Comput Sci Netw Secur 2017.

Aboobucker

, Bao

What obstruct customer acceptance of internet banking? Security and privacy, risk, trust and website usability and the role of moderators, J High Technol Manag Res 2018. doi: 10.1016/j.hitech.2018.04.010.

Pikkarainen

, Pikkarainen

, Karjaluoto

, Pahnila

Consumer acceptance of online banking: An extension of the technology acceptance model, Internet Research 2004. doi: 10.1108/10662240410542652.

Almaiah

M. A.

et al., Factors influencing the adoption of internet banking: An integration of ISSM and UTAUT with price value and perceived risk, Front Psychol 2022. doi: 10.3389/fpsyg.2022.919198.

Shih

Y.Y.

, Fang

The use of a decomposed theory of planned behavior to study Internet banking in Taiwan, Internet Res 2004. doi: 10.1108/10662240410542643.

Miranda-petronella

E-Banking- Modern Banking Services, Ann Univ Oradea –Econ Sci 2009.

Boateng

, Adam

D.R.

, Okoe

A.F.

, Anning-Dorson

Assessing the determinants of internet banking adoption intentions: A social cognitive theory perspective, Comput Human Behav 2016. doi: 10.1016/j.chb.2016.09.017.

Khalfan

A.M.S.

, AlRefaei

Y.S.Y.

, Al Hajery

Factors influencing the adoption of internet banking in Oman: a descriptive case study analysis, Int J Financ Serv Manag 2006. doi: 10.1504/ijfsm.2006.009623.

Zahid

, Mujtaba

, Riaz

Consumer acceptance of online banking, Eur J Econ Financ Adm Sci 2010.

10.

Rahi

, Abd.Ghani

, Hafaz Ngah

Integration of unified theory of acceptance and use of technology in internet banking adoption setting: Evidence from Pakistan, Technol Soc 2019. doi: 10.1016/j.techsoc.2019.03.003.

11.

Montazemi

A.R.

, Qahri-Saremi

Factors affecting adoption of online banking: A meta-analytic structural equation modeling study, Inf Manag 2015. doi: 10.1016/j.im.2014.11.002.

12.

George

Perceptions of Internet banking users –a structural equation modelling (SEM) approach, IIMB Manag Rev 2018. doi: 10.1016/j.iimb.2018.05.007.

13.

Abu-Taieh

E.M.

et al., Continued Intention to Use of MBanking in Jordan by Integrating UTAUT, TPB, TAM and Service Quality with ML, J Open Innov Technol Mark Complex 2022. doi: 10.3390/joitmc8030120.

14.

Maduku

D.K.

Customers’ adoption and use of e-banking services: The South African perspective, Banks Bank Syst 2014.

15.

Alhassany

, Faisal

Factors influencing the internet banking adoption decision in North Cyprus: an evidence from the partial least square approach of the structural equation modeling, Financ Innov 2018. doi: 10.1186/s40854-018-0111-3.

16.

Carranza

, Díaz

, Sanchez-Camacho

, Martín-Consuegra

e-Banking Adoption: An Opportunity for Customer Value Co-creation, Front Psychol 2021. doi: 10.3389/fpsyg.2020.621248.

17.

Oliveira

, Thomas

, Baptista

, Campos

Mobile payment: Understanding the determinants of customer adoption and intention to recommend the technology, Comput Human Behav 2016. doi: 10.1016/j.chb.2016.03.030.

18.

Qendraj

D.H.

, Xhafaj

, Halidini

Ranking the Most Important Attributes of using Google Classroom in online teaching for Albanian Universities: A Fuzzy AHP Method with Triangular Fuzzy Numbers and Trapezoidal Fuzzy Numbers, Adv Sci Technol Eng Syst J 2021. doi: 10.25046/aj060134.

19.

Halidini

et al., Assessing the university online learning adapted with UTAUT2 model: A fuzzy z-AHP approach, Adv Math Sci J 2022. doi: 10.37418/amsj.11.4.5.

20.

Qendraj

D.H.

, Xhafaj

, Thanasi

A hybridized approach of PLS-SEM and fuzzy Z-AHP to evaluate the UTAUT2 model for an LMS, J Intell Fuzzy Syst 2022. doi: 10.3233/JIFS-212396.

21.

Kosova

, Qendraj

D.H.

, Xhafaj

Meta-Analysis ELECTRE III and AHP in Evaluating and Ranking the Urban Resilience, J Environ Manag Tour 2022. doi: 10.14505/jemt.v13.3(59).15.

22.

Lorena

A.C.

, Carvalho

A.C.P.L.F.

Evolutionary design of multiclass support vector machines, 2007.

23.

Xhafaj

, Qendraj

D.H.

, Xhafaj

, Gjikaj

A Hybrid Integration of PLS-SEM, AHP, and FAHP Methods to Evaluate the Factors That Influence the Use of an LMS, Int J Decis Support Syst Technol 2021. doi: 10.4018/ijdsst.286697.

24.

Fishbein

, Ajzen

Strategies of Change: Active Participation, Belief, attitude, intention, and behavior: An introduction to theory and research, 1975.

25.

Ajzen

, Ajzen

The theory of planned behavior.Organizational Behavior and Human Decision Processes.The theory of planned behavior, Organizational Behavior and Human Decision Processes 1991.

26.

Davis

F.D.

, Bagozzi

R.P.

, Warshaw

P.R.

User Acceptance of Computer Technology: A Comparison of Two Theoretical Models, Manage Sci 1989. doi: 10.1287/mnsc.35.8.982.

27.

Ajzen

The theory of planned behavior, Organ Behav Hum Decis, Process 1991.

28.

Venkatesh

, Morris

M.G.

, Davis

G.B.

, Davis

F.D.

User acceptance of information technology: Toward a unified view, MIS Q Manag Inf Syst 2003. doi: 10.2307/30036540.

29.

Martins

, Oliveira

, Popovic

Understanding the internet banking adoption: A unified theory of acceptance and use of technology and perceived risk application, Int J Inf Manage 2014. doi: 10.1016/j.ijinfomgt.2013.06.002.

30.

Nor

K.M.

, Sutanonpaiboon

, Mastor

N.H.

Malay, Chinese, and internet banking, Chinese Manag Stud 2010. doi: 10.1108/17506141011053069.

31.

Zhou

, Lu

, Wang

Integrating TTF and UTAUT to explain mobile banking user adoption, Comput Human Behav 2010. doi: 10.1016/j.chb.2010.01.013.

32.

Riffai

M.M.M.A.

, Grant

, Edgar

Big TAM in Oman: Exploring the promise of on-line banking, its adoption by customers and the challenges of banking in Oman, Int J Inf Manage 2012. doi: 10.1016/j.ijinfomgt.2011.11.007.

33.

Abbad

M.M.

E-banking in Jordan, Behav Inf Technol 2013. doi: 10.1080/0144929X.2011.586725.

34.

Rahi

, Othman Mansour

M.M.

, Alghizzawi

, Alnaser

F.M.

Integration of UTAUT model in internet banking adoption context: The mediating role of performance expectancy and effort expectancy, J Res Interact Mark 2019. doi: 10.1108/JRIM-02-2018-0032.

35.

Vuković

, Pivac

, Kundid

Technology Acceptance Model for the Internet Banking Acceptance in Split, Bus Syst Res 2019. doi: 10.2478/bsrj-2019-022.

36.

Xhafaj

, Qendraj

D.H.

, Xhafaj

, Halidini

Analysis and evaluation of factors affecting the use of google classroom in Albania: A partial least squares structural equation modelling approach, Math Stat 2021. doi: 10.13189/ms.2021.090205.

37.

Chaouali

, Ben

The interplay of counter-conformity motivation, social influence, and trust in customers’ intention to adopt Internet banking services: The case of an emerging country, J Retail Consum Serv 2016. doi: 10.1016/j.jretconser.2015.10.007.

38.

Lin

F.T.

, Wu

H.Y.

, Tran

T.N.N.

Internet banking adoption in a developing country: an empirical study in Vietnam, Inf Syst E-bus Manag 2015. doi: 10.1007/s10257-014-0268-x.

39.

Gupta

, Arora

Investigating consumer intention to accept mobile payment systems through unified theory of acceptance model: An Indian perspective, South Asian J Bus Stud 2020. doi: 10.1108/SAJBS-03-2019-0037.

40.

Venkatesh

, Morris

, Davis

Technology acceptance model –research, MIS Q 2003.

41.

Hong

I.B.

Understanding and predicting behavioral intention to adopt mobile banking: The Korean experience, J Glob Inf Manag 2019. doi: 10.4018/JGIM.2019070110.

42.

Farzin

, Sadeghi

, Yahyayi

Extending UTAUT2 in M-banking adoption and actual use behavior: Does WOM communication matter? Asian J Econ Bank 2021. doi: 10.1108/ajeb-10-2020-0085.

43.

Salisbury

W.D.

, Pearson

R.A.

, Pearson

A.W.

, Miller

D.W.

Perceived security and World Wide Web purchase intention, Ind Manag Data Syst 2001. doi: 10.1108/02635570110390071.

44.

Shin

D.H.

Towards an understanding of the consumer acceptance of mobile wallet, Comput Human Behav 2009. 10.1016/j.chb.2009.06.001.

45.

C.T.B.

, Lin

W.C.

Measuring the service quality of internet banking: Scale development and validation, Eur Bus Rev 2010. doi: 10.1108/09555341011008981.

46.

Blut

E-Service Quality: Development of a Hierarchical Model, J Retail 2016. doi: 10.1016/j.jretai.2016.09.002.

47.

Cristobal

, Flavian

, Guinalíu

Perceived eservice quality (PeSQ): Measurement validation and effects on consumer satisfaction and web site loyalty, Manag Serv Qual 2007. doi: 10.1108/09604520710744326.

48.

Al-Qeisi

, Dennis

, Alamanos

, Jayawardhena

Website design quality and usage behavior: Unified theory of acceptance and use of technology, J Bus Res 2014. doi: 10.1016/j.jbusres.2014.06.016.

49.

Swaid

S.I.

, Wigand

R.T.

Key dimensions of e-commerce service quality and its relationships to satisfaction and loyalty, 2007.

50.

Rahi

, Abd

The influence of e-customer services and perceived value on brand loyalty of banks and internet banking adoption: A structural equation model (SEM), J Internet Bank Commer 2017.

51.

Hair

et al., A Primer on Partial Least Square Structural Equation Modelling (PLS-SEM). 2017.

52.

Benitez

, Henseler

, Castillo

, Schuberth

How to perform and report an impactful analysis using partial least squares: Guidelines for confirmatory and explanatory IS research, Inf Manag 2020. doi: 10.1016/j.im.2019.05.003.

53.

Hair

J.F.

, Risher

J.J.

, Sarstedt

, Ringle

C.M.

When to use and how to report the results of PLS-SEM, European Business Review 2019. doi: 10.1108/EBR-11-2018-0203.

54.

Dijkstra

T.K.

, Henseler

Consistent partial least squares path modeling, MIS Quarterly: Management Information Systems 2015. doi: 10.25300/MISQ/2015/39.2.02.

55.

Haykin

Neural networks: a comprehensive foundation by Simon Haykin, The Knowledge Engineering Review 1999.

56.

Masters

Practical Neural Networks Recipes in C++, 1993.

57.

Sharma

S.K.

, Sharma

Examining the role of trust and quality dimensions in the actual usage of mobile banking services: An empirical investigation, Int J Inf Manage 2019. doi: 10.1016/j.ijinfomgt.2018.09.013.

58.

Chong

A.Y.L.

Atwo-staged SEM-neural network approach for understanding and predicting the determinants of m-commerce adoption, Expert Syst Appl 2013. doi: 10.1016/j.eswa.2012.08.067.

59.

Al-Shihi

, Sharma

S.K.

, Sarrab

Neural network approach to predict mobile learning acceptance, Educ Inf Technol 2018. doi: 10.1007/s10639-018-9691-9.

60.

Anouze

A.L.M.

, Alamro

A.S.

Factors affecting intention to use e-banking in Jordan, Int J Bank Mark 2020. doi: 10.1108/IJBM-10-2018-0271.

61.

Lee

V.H.

, Hew

J.J.

, Leong

L.Y.

, Tan

G.W.H.

, Ooi

K.B.

Wearable payment: A deep learning-based dualstage SEM-ANN analysis, Expert Syst Appl 2020. doi: 10.1016/j.eswa.2020.113477.

62.

Sternad Zabukovšek

, Kalinic

, Bobek

, Tominc

SEM–ANN based research of factors’ impact on extended use of ERP systems, Cent Eur J Oper Res 2019. doi: 10.1007/s10100-018-0592-1.

63.

Garson

G.D.

A Comparison of Neural Network and Expert Systems Algorithms with Common Multivariate Procedures for Analysis of Social Science Data, Soc Sci Comput Rev 1991. doi: 10.1177/089443939100900304.

64.

Leong

L.Y.

, Hew

T.S.

, Tan

G.W.H.

, Ooi

K.B.

Predicting the determinants of the NFC-enabled mobile credit card acceptance: A neural networks approach, Expert Syst Appl 2013. doi: 10.1016/j.eswa.2013.04.018.

65.

Hwang

C.-L.

, Yoon

Multiple Attribute Decision Making Methods and Applications A State-of-the-Art Survey, Lect Notes Econ Math Syst 1981.

66.

Sun

, Zhang

, Ni

AHybrid Model Based onSEM and Fuzzy TOPSIS for Supplier Selection, Mathematics 2022. doi: 10.3390/math10193505.

Constructs	Loadings	α	CR	AVE
and Items
(BI)		0.866	0.918	0.789
BI1	0.89
BI2	0.901
BI3	0.872
(CS)		0.841	0.894	0.678
CS1	0.759
CS2	0.815
CS3	0.856
CS4	0.86
(EE)		0.873	0.914	0.726
E1	0.855
E2	0.88
E3	0.87
E4	0.801
(FC)		0.75	0.856	0.665
FC1	0.83
FC2	0.847
FC3	0.766
(SI)		0.891	0.932	0.82
SI1	0.885
SI2	0.873
SI3	0.887
(HT)		0.86	0.905	0.705
HT1	0.876
HT2	0.875
HT3	0761
HT4	0.841
(PE)		0.853	0.9	0.693
PE1	0.818
PE2	0.849
PE3	0.835
PE4	0.828
(PS)		0.893	0.933	0.823
PS1	0.889
PS2	0.908
PS3	0.924
(UB)		0.87	0.92	0.794
UB1	0.877
UB2	0.881
UB3	0.915