Using FAHP and CBR to evaluate the intention of adoption of internet banking service: The example of web ATM

Abstract

This research combines the Fuzzy Analytic Hierarchy Process (FAHP) with Case-Based Reasoning (CBR) to evaluate the intention of adoption of web ATM services. Compared with physical ATM service, web ATM allows users to perform financial transactions over the internet conveniently. Based on literature and considering the characteristics of web ATM, this study constructs a model for web ATM adoption that comprises three dimensions: The knowledge, the potential value, and the security. 222 valid user questionnaires are collected, and factor analysis is used to verify the factor structure of the decision hierarchy. FAHP is then used to calculate the weights of criteria with six experts through pairwise comparisons. Finally, FAHP weights are integrated into a CBR prediction mechanism for evaluating a user’s adoption intention toward web ATM. The results are helpful for financial institutions to understand and to evaluate the user behavior toward internet banking service adoption.

Keywords

Fuzzy analytic hierarchy process case-based reasoning web ATM innovation adoption

1 Introduction

The financial institutions in Taiwan have focused on developing internet banking, which is not limited to location-based banking services or self-service terminals, such as ATM. Innovative information services have increased consumers’ choice of methods for conducting financial transactions. When people are increasingly relying on the internet, internet-based transactions including e-auctions, e-cash, e-payments, and e-bidding are becoming more diversified and convenient for the users. Therefore, the multi-channel strategy has become the most popular business model for banks [15, 43].

According to a MIC online-shopping consumer survey in 2016, the most frequently used payment method is credit cards (75.7%), followed by the collection at convenience stores (66.4%), and automatic teller machines (ATM) payment (25.2%) [28]. ATM payments can be made via physical ATM by using ATM cards with a fund transfer function, or consumers can use online transfer via web ATM with a card reader. A web ATM is an online interface for ATM card banking that provides all the usual ATM functions without withdrawing cash. According to the InsightXploer survey, netizens in Taiwan use ATM (11.5%) vs. web ATM (7.8%) as a payment method for online shopping in 2016 [17].

Although most banks in Taiwan provide these online services, what are the factors influencing the decision of user adoption intention toward web ATM services? While many prior studies have investigated the factors influencing user acceptance of Internet banking [7 , 47], little research has been conducted on the factor affecting user adoption of Web ATM. Thus, the objective of this study is to identify and evaluate the decision factors and to predict web ATM adoption. It is hoped that the results of this study will contribute to our understanding of the user decision toward web ATM adoption. The factors examined here could serve as the basis for a study in explaining the adoption of internet banking services.

The article is structured as follows. The first section deals with the theoretical foundations for the development of the research model of web ATM adoption based on the innovation adoption literature. After which research methodology is presented, with user questionnaire and factor analysis used to verify the factor structure of the decision hierarchy. FAHP is then used to calculate the weights of criteria, which are applied to the CBR prediction mechanism to evaluate the user intention toward web ATM adoption. Finally, results are discussed and conclusions are drawn.

2 Web ATM adoption model

The world’s first cash machine landed installed on a north London of Barclays branch in Enfield, transforming everyday banking for millions of people all over the globe [42], the use of cash dispensers (CDs) and ATM spread rapidly across the world. However, Taiwan has one Automated Teller Machine (ATM) per 826 residents, giving the island country the highest ATM density in the world [38].

In a content analysis of innovation definitions, “Innovation is the multi-stage process whereby organizations transform ideas into new/improved products, service or processes, in order to advance, compete and differentiate themselves successfully in their marketplace [2].” Evolved from ATM, web ATM is an Internet-based ATM is an online ATM terminal that does not give out cash bills. the flexibility represented by an ATM operating over the Internet, with instantaneous customer personalization, payment options, pricing alternatives and similar customized feature/functionality can be combined with speed, flexibility and lower cost of operational support, the Web ATM may prove to be a key ingredient in realizing the long term potential of the lowly “banker in a box” [1].

As the online banking services of various financial institutions are continually being updated, web ATM provide more personalized services for cardholders than physical ATM, such as common account editing, transaction message notification, and et al. Web ATM offer Internet-based banking services, for which user adoption of web ATM and their attitudes toward the use of web ATM can be seen as an innovation adoption. Brand and Huizingh proposed five factors that influence future intention to adopt e-commerce: knowledge, potential value, implementation, satisfaction, and adoption level [4]. Among them, knowledge refers to the e-commerce knowledge that a company possesses, such as knowledge of technology and customer needs of e-commerce. Potential value involves two aspects, the potential benefits and the costs of innovation adoption, and is similar to the concepts of relative advantage and perceived benefits. Adoption level refers to the level of e-commerce use in sales, purchasing, and customer service. Implementation refers to the complete integration and use of existing e-commerce software and hardware. Satisfaction refers to the results that emerge from the potential value variable of e-commerce.

This study develops a research model for user intentions regarding web ATM adoption. Based on the adoption model proposed by Brand and Huizingh, the knowledge and potential value dimensions are included in this study; whereas satisfaction, fulfillment, and adoption level are excluded because the target subjects include potential users who have not previously used web ATM [4]. This study also includes the security dimension which refers to the elimination of threats, risk, and suspicion through methods including physical security, financial security, and privacy protection in the model of service quality proposed by Zeithaml et al. [49]. Besides, Sukkar and Hasan reported that customers in developing countries are concerned about the security issues for the acceptance of internet banking [36]. This concern may affect their willingness to use Internet banking services, and reasons for not using internet banking services include a lack of customer trust, sound government policies, and regulatory protection. In summary, this study proposed three dimensions in the decision hierarchy of web ATM adoption: the knowledge, the potential value, and the security.

The knowledge criteria were primarily established to determine whether users possess sufficient knowledge of the hardware and software and whether users are familiar with the software operation. The knowledge criteria include knowledge regarding security component installation, installation of chip reader, the specification of required hardware, the need for browser version or above, and high compatibility.

The potential value criteria primarily emphasize user recognition of the value and potential benefits of web ATM. The potential value criteria include preferable handling fees, time savings, greater service capability, higher selectivity, no time restrictions, and easier financial management compared to physical ATM. The security criteria primarily emphasize the security considerations concerning transaction activities when using web ATM. So the security criteria include criteria such as leakage of transaction data, transaction errors, computer viruses, authentication method, and safety of transaction platform.

3 Research method

3.1 Research process

The research process is shown in Fig. 1. First, based on the literature, the research model of web ATM adoption is developed with three dimensions. The questionnaire items for general ATM users are developed based on the criteria in the model. 222 valid user samples are collected. Factor analysis is then used to verify the factor structure of the decision hierarchy. Second, this study adopts FAHP to constructs matrices for calculating a set of expert pairwise comparison. Six expert responses were collected and used to calculate the relative FAHP weight of the criteria to each level, and then structures the decision hierarchy as Fig. 2. Finally, this study employs Visual Studio.Net as the development tool to develop a prototype combining FAHP weights with CBR prediction.

Fig. 1

The research process.

Fig. 2

The research model of web ATM adoption.

3.2 AHP and FAHP

The AHP was developed by Saaty in the 1970 s to define priorities and support decision making. To deal with the complexity of decision-making, the factors that affect the decision need to be identified and organized in a hierarchical structure. It is a method of measurement through pairwise comparisons which rely on the judgments of experts to determine priority scales to show how many times more important one element is over another element with respect to the criterion [32].

Alternative methods are reported such as Analytic Network Process (ANP), Technique for Order of Preference by Similarity to Ideal Solution (TOPSIS) and et al. [31]. ANP is an extension of AHP which adopts a network structure to present the relationship between criteria and alternatives. Reisi et al. applied ANP and AHP to the decision for industrial site selection [30]. Lin et al. applying the ANP and linear programming (LP) methods to provide the solution for green supplier selection and resolve the problem of order allocation for each vendor [23]. TOPSIS identifies solutions from a set of alternatives, in which the geometric distance between each alternative and the ideal alternative is calculated. Zyoud and Fuchs-Hanusch conducted a bibliometric survey on AHP and TOPSIS researches [50].

The AHP method can be extended and integrated with different techniques. Fuzzy theory combined with AHP to form fuzzy AHP or FAHP [44]. To deal with the vagueness of human thought, Zadeh first introduced the fuzzy set theory [48], which was oriented to the rationality of uncertainty due to imprecision or vagueness. A fuzzy set is a class of objects with a continuum of grades of membership. Such a set is characterized by a membership (characteristic) function, which assigns to each object a grade of membership ranging between zero and one [3]. A tilde “∼” is placed above a symbol if the symbol represents a fuzzy set. Therefore, $\tilde{p}, \tilde{r}, \tilde{n}$ are all fuzzy sets. The membership functions for these fuzzy sets are denoted by $μ (x | \tilde{p})$ , and $μ (x | \tilde{n})$ respectively. However, in general, it is a rather subjective way to determine the form of a fuzzy number corresponding to the semantic term [26]. Three popular fuzzy numbers are often used to analyze data. They are triangular fuzzy number, trapezoidal fuzzy number, and normal fuzzy number. They are frequently used to quantification linguistic variables [22]. In this study, take the 5-point Likert scale as an example, the measurement of its semantic variable is expressed as a symmetric triangular fuzzy number, and its membership function is shown in Fig. 3.

Fig. 3

Fuzzy semantic variables based on Likert.

A triangular fuzzy number (TFN), $\tilde{M}$ , is shown in Fig. 4. A TFN is denoted simply as (m1/m2, m2/m3) or (m1, m2, m3). The parameters m1, m2, and m3 respectively denote the smallest possible value, the most promising value, and the largest possible value that describe a fuzzy event [18].

Fig. 4

A triangular fuzzy number, $\tilde{M}$ .

Each TFN has linear representations on its left and right side such that its membership function can be defined as: $μ (x | \tilde{M}) = {\begin{matrix} \begin{matrix} \begin{matrix} 0 & , x < m_{1} \end{matrix} \\ \begin{matrix} (x - m_{1}) / (m_{2} - m_{1}) & , m_{1} \leq x \leq m_{2} \end{matrix} \\ \begin{matrix} (m_{3} - x) / (m_{3} - m_{2}) & , m_{2} \leq x \leq m_{3} \end{matrix} \end{matrix} \\ \begin{matrix} 0 & , x > m_{3} \end{matrix} \end{matrix}$ (1)

A fuzzy number can always be given by its corresponding left and right representation of each degree of membership: $\begin{matrix} \tilde{M} = (M^{l (y)}, M^{r (y)}) = (m_{1} + (m_{2} - M_{1}) y, m_{3} \\ + (m_{2} - m_{3}) y, y ɛ [0, 1]) \end{matrix}$ (2) where l(y) and r(y) denote the left side representation and the right side representation of a fuzzy number respectively.

According to Brasil et al. [5], “Many decision-making and problem-solving tasks are too complex to be understood quantitatively, however, people succeed by using knowledge that is imprecise rather than precise.” Fuzzy set theory mirrors personal reasoning in its use of approximate information and uncertainty to make the human-like decision making. Fuzzy set theory encompasses fuzzy logic, fuzzy arithmetic, fuzzy mathematical programming, fuzzy topology, fuzzy graph theory, and fuzzy data analysis. However, the term fuzzy logic is often used to describe all of these.

The analytic hierarchy process (AHP) is one of the extensively used multi-criteria decision-making methods. One of the main advantages of this method is the relative ease with which it handles multiple criteria. In addition to this, AHP is easier to understand, and it can effectively handle both qualitative and quantitative data. AHP involves the principles of decomposition, pair-wise comparisons, and priority vector generation and synthesis. Though the purpose of AHP is to capture the expert’s knowledge, the conventional AHP still cannot reflect the human thinking style. Therefore, fuzzy AHP, a fuzzy extension of AHP, was developed to solve the hierarchical fuzzy problems [14].

Samuel et al. used FAHP technique to compute the weights of the attributes which were applied to train an Artificial Neural Network (ANN) classifier to predict heart failure risks in patients [35]. Tavana et al. integrated AHP, Adaptive Neuro Fuzzy Inference System (ANFIS), and ANN to predict and rank the suppliers’ performance [41]. FAHP is widely used by researchers to present a variety of problems and to evaluate criteria. There are systematic reviews of FAHP applications [8 , 39]. Researchers mainly used literature to build the criteria, and FAHP to calculate their weights [31].

This study adopts FAHP to calculate criteria weights. The steps of generating FAHP weights are as follows: the positive reciprocal matrix was converted into a fuzzy positive reciprocal matrix; followed by group integration; then the fuzzy weights were calculated; and finally, the data were defuzzified and normalized.

First, the assessed value of the relative importance between the two criteria judged by experts using semantic variables was compiled according to group judgments. These semantic variables can be expressed using a triangular fuzzy number, which then uses the mean to integrate the judges of various experts as shown in Equation 3. ${\tilde{T}}_{ij} = 1 / K ({\tilde{t}}_{ij}^{1} + {\tilde{t}}_{ij}^{2} + \dots + {\tilde{t}}_{ij}^{k}) = (t_{ij}^{l}, t_{ij}^{m}, t_{ij}^{u})$ (3) where ${\tilde{T}}_{ij}$ is the importance comparison value of the i^th and j^th criteria after the integration of K number of expert advice; ${\tilde{t}}_{ij}^{k}$ is the importance comparison value of the kth expert on the i^th and j^th criteria; and K is the total number of experts.

After establishing the positive reciprocal matrix by integrating all the experts’ judgments, as shown in Equation 4 to Equation 6, a fuzzy positive reciprocal matrix can be established, where T is the fuzzy positive reciprocal matrix. $T = [{\tilde{T}}_{ij}]$ (4) ${\tilde{T}}_{ij} = 1, \forall i = j$ (5) ${\tilde{T}}_{ij} = 1 / {\tilde{T}}_{ji}, i \neq j, \forall i, j = 1, 2, \dots, n$ (6)

According to the fuzzy weights calculated using the fuzzy positive reciprocal matrix, as indicated in a study by Buckley, the calculation of fuzzy weight is accomplished using the row vector of geometric mean; which can not only obtain the fuzzy weight of the fuzzy positive reciprocal matrix, but can also achieve normalization as an objective [6]. Fuzzy weight is calculated as shown in Equation 7 and Equation 8. ${\tilde{r}}_{i} = ({\tilde{a}}_{i 1} \times {\tilde{a}}_{i 2} \times \dots \times {\tilde{a}}_{in})^{1 / n}$ (7)

Where i is the row number, n is the column number (the assessment index number of the layer), ${\tilde{a}}_{ij}$ is the i^th row and j^th column of the triangular fuzzy number in the fuzzy positive reciprocal matrix. ${\tilde{w}}_{i} = {\tilde{r}}_{i} \times ({\tilde{r}}_{1} + {\tilde{r}}_{2} + \dots + {\tilde{r}}_{n})^{- 1}$ (8)

Where ${\tilde{r}}_{i}$ is the geometric mean of a triangular fuzzy number, ${\tilde{w}}_{i}$ is the i^th fuzzy weight factor.

Since these calculated weights are fuzzy weights, they need to go through the defuzzification process to obtain the critical success factors or these weights. When the weight of the triangular fuzzy number is ${\tilde{A}}_{ij} = (L_{ij}, M_{ij}, U_{ij})$ , the calculation for the fuzzy weight dF_ij and the normalized NW_i is listed in Equation 9 and Equation 10. ${dF}_{ij} = \frac{(U_{ij} - L_{ij}) + (M_{ij} - L_{ij})}{3} + L_{ij}$ (9) ${NW}_{i} = \frac{{dF}_{ij}}{\sum {dF}_{ij}}$ (10)

3.3 CBR prediction with FAHP weights

CBR is a computer reasoning method that is consistent with the analogical decision making in everyday human problem-solving. CBR use past experiences to address new situations [20]. CBR defines the description of the problem as a new case at the beginning. To address a new situation, the new case is compared with older cases in the case base to retrieve the most similar ones and find a suggested solution among them [20]. The solution is later validated through feedback from the user or the environment. Finally, the validated solution can be retained in the case base for use in future problem-solving if appropriate.

Research have applied CBR to a variety of domains ranging from pattern classification [46], to cost estimate [29], identifying knowledge leader within online community [37], diagnostic decision support system for healthcare-associated infections [12]. To design an appropriate case-matching mechanism in the retrieval stage, Yan et al. proposed using the learning pseudo metrics (LPM) to replace the widely used Euclidean distance in similarity function [46]. A similarity function or functions are used to compute the degree of match between the input case and the target case. The LPM-based retrieval method has been demonstrated improving the quality and learning ability of CBR classifiers.

Instead of focusing on measuring the degree of feature match in similarity functions, this study determines a set of expert weights for the case features. In a case-based approach representing a case with features tied to a context at an operational level is an important issue. Every feature in the input case is matched to a corresponding feature in the retrieved case. A weight is usually assigned to each case feature representing the importance of that feature to the match.

Hsu et al. pointed out that the feature weights can be assigned to a set of prior known fixed values or all set equal if no arbitrary priorities are determined. Usually, cases with higher degrees of match are retrieved and adapted to generate a solution [16]. However, the solution cannot be guaranteed if the weights are determined using arbitrary human judgment. A systematic mechanism for determining a set of expert weights can improve case retrieval effectiveness. Qin et al. found that CBR was an effective method in cost estimating and that AHP weights are better than equal weight and gradient descent method [29]. Wu et al. proposed a fuzzy CBR retrieving mechanism in which FAHP was used to determine weights for retrieve product-ideas that tend to enhance the functions of the baseline product [45].

The CBR prediction using FAHP weights can be considered as a four-step reasoning process. First, the new case is inputted into the prototype system. Second, CBR uses the similarity function to locate cases in the case base that are most similar to the input case. The new case is compared with each case in the case base to calculate the degree of similarity based on the FAHP weights. Third, the top k number of similar cases are retrieved based on the degree of similarity. Finally, the retrieved cases are adapted to generate the prediction result for the input case.

In web ATM adoption context, the case base is consisted of the user sample. A nearest-neighbor matching function containing the weights in the formula is shown in Equation 9. $Similarity (f^{I}, f_{j}^{R}) = \frac{\sum_{i = 1}^{l} W_{i} \times Sim (f_{i}^{I}, f_{ji}^{R})}{\sum_{i = 1}^{l} W_{i}}$ (11)

Where f^I is the input case; $f_{j}^{R}$ is the j^th case in the case base; $Similarity (f^{I}, f_{j}^{R})$ is of similarity between the input case and the j^th case in the case base; W_i is the i^th weight attribute; $f_{i}^{I}$ is the i^th attribute value of the input case; $f_{ij}^{R}$ is the i^th attribute value of the j^th case in the case base; l is the number of attributes; $Sim (f_{i}^{I}, f_{ji}^{R})$ is the degree of similarity between $f_{i}^{I}$ and $f_{ji}^{R}$ , which is designed as one minus the absolute standardized difference value: 1-difference value / maximum difference value.

4 Data collection and analysis

A survey instrument was used to assess the factors and verify the proposed model. We first focus on measurement, the criteria reflecting the factors in the model were generated based on related literature and web ATM characteristics. A 5-point Likert scale was used for all measures to provide consistency for responses. The scales ranged from 1 indicating “Strongly Disagree” to 5 denoting “Strongly Agree.” In preparation for large-scale data collection, the survey instrument was pretested. These items were first reviewed for content by one experienced web ATM expert, one web ATM user, and one physical ATM user without the usage experience of web ATM. The respondents were asked to report encountered problems regarding the comprehensiveness of the instrument. The instrument was modified based on this feedback.

The revised questionnaire with the refined measurement was then used for a survey. A total of 248 surveys were delivered. The completeness of each received questionnaire was carefully examined. 26 questionnaires were thus dropped resulting in 222 valid ones, yielding a valid response rate of 89.52%. The descriptive respondent characteristic statistics are shown in Table 1. Ninety-one of them have experience using web ATM.

Table 1
Descriptive statistics of respondents’ characteristics (n = 222)

Characteristic Range Frequency Percentage(%) Cumulative Percentage(%)

Male 129 58 58

Gender Female 93 42 100

Background Military, civil servant, teacher 53 24 24

Information and communication 41 18 42

Student 37 17 59

Service, finance, and insurance 35 16 75

Manufacturing and construction 19 9 83

Freelance 11 5 88

Other 26 12 100

Age Under 20 20 9 9

21 – 30 123 55 64

31 – 40 64 29 93

More than 41 15 7 100

Online shopping experience Yes 169 76 76

No 53 24 100

Online auction experience Yes 152 68 68

No 70 32 100

Web ATM experience Yes 91 41 41

No 131 59 100

Characteristic	Range	Frequency	Percentage(%)	Cumulative Percentage(%)
	Male	129	58	58
Gender	Female	93	42	100
Background	Military, civil servant, teacher	53	24	24
	Information and communication	41	18	42
	Student	37	17	59
	Service, finance, and insurance	35	16	75
	Manufacturing and construction	19	9	83
	Freelance	11	5	88
	Other	26	12	100
Age	Under 20	20	9	9
	21 – 30	123	55	64
	31 – 40	64	29	93
	More than 41	15	7	100
Online shopping experience	Yes	169	76	76
	No	53	24	100
Online auction experience	Yes	152	68	68
	No	70	32	100
Web ATM experience	Yes	91	41	41
	No	131	59	100

The gender and background distributions were as follows: ninety-three respondents were female (42%) and 129 respondents were male (58%); 53 respondents were from the military, civil servant, and teacher background (24%), followed by 41 respondents from the information and communication industry (18%) and 37 student respondents (17%); 169 respondents had online shopping experience (76%); 152 respondents had online auction experience (68%); and 91 respondents had experience using web ATM (41%).

The sample was used for exploratory factor analysis to identify the factor structure hidden in the data collected. SPSS was used for the analysis. All the 16 items were used to run factor analysis; principal components were adopted, and varimax was used to turn the axle. Kaiser-Meyer-Olkin (KMO) test is a measure of how suited the data is for factor analysis. The test measures sampling adequacy for each criterion in the model and for the complete model. As shown in Table 2, the KMO measure is 0.886 (>0.8), which is considered meritorious for conducting a factor analysis with the data [19].

Table 2

Results of factor analysis (KMO:0.886)

Dimension/Criteria	Sub-criteria	Factor 1	Factor 2	Factor 3
Potential value (α=0.915)	P6	0.831	0.271	0.051
	P5	0.819	0.258	0.072
	P4	0.811	0.246	0.148
	P2	0.809	0.163	0.185
	P1	0.795	0.293	0.098
	P3	0.742	0.084	0.131
Knowledge (α=0.947)	K3	0.200	0.892	0.158
	K4	0.182	0.864	0.230
	K1	0.284	0.843	0.163
	K2	0.267	0.811	0.178
	K5	0.312	0.799	0.221
Security (α=0.930)	S1	0.020	0.187	0.883
	S3	0.037	0.220	0.877
	S2	0.064	0.225	0.849
	S5	0.230	0.141	0.806
	S4	0.292	0.072	0.764
Eigenvalue	7.522	2.733	1.875
Variance %	27.023	25.177	23.612
Cumulative Variance %	27.023	52.200	75.812

According to the eigenvalue rule, a three-factor structure emerged. Those factors with eigenvalues greater than one were selected. Criterion with factor loading higher than 0.50 was used for the identification of the factor [13]. All criteria in the model were loaded on their respective factors, and no criterion loaded higher on factors they were not intended to measure.

Reliability of each factor was assessed by Cronbach’s α. There are different reports suggested the acceptable values of alpha, ranging from 0.70 to 0.95 [40]. As shown in Table 2, the value of alpha associated with each factor is at an acceptable level, ranging from 0.915 to 0.947. In addition, the factors are listed in the order of the variance they explain. The first factor, potential value, explains 27.023% of the variance in the data; the second factor is knowledge (25.177%); the third factor is security (23.612%). The results of factor analysis support the three-factor structure proposed in the research model.

Subsequently, FAHP methods was used for calculating the weights for CBR prediction. Six expert responses were collected from one senior manager and one manager in online banking department, one clerk handling ATM services in customer service department at banks; one R&D department manager, one manager and one assistant manager in IT department at ATM service companies. They are inquired to determine priority scales through pairwise comparisons.

The Expert Choice and Microsoft Excel software was adopted to obtain the FAHP weights using the judgments of six experts. Table 3 shows the resulted FAHP weights in the criteria level. The consistency index (CI) value of criteria is 0.01 (< =0.1), and the consistency ratio (CR) value is 0.02 (< =0.1) which indicated the expert judgments were consistent [33].

Table 3

FAHP weights

Dimension/Criteria	Weight	Sub-criteria	Weight	Global Weight	Rank
Security	0.770	S1	0.331	0.255	1
		S2	0.229	0.176	2
		S3	0.186	0.143	3
		S4	0.149	0.115	4
		S5	0.106	0.082	5
Knowledge	0.125	K1	0.413	0.052	6
		K2	0.175	0.022	8
		K3	0.140	0.018	12
		K4	0.154	0.019	9
		K5	0.118	0.015	13
		P1	0.275	0.029	7
Potential Value	0.105	P2	0.180	0.019	10
		P3	0.172	0.018	11
		P4	0.137	0.014	14
		P5	0.135	0.014	15
		P6	0.101	0.011	16

The mean average errors (MAE) for CBR prediction are shown in Table 4. It is indicated that FAHP+CBR outperforms AHP+CBR in all the MAE results for k = 1 to 10. Also, the average MAE of FAHP+CBR is 0.40 and AHP+CBR 0.44. The CBR predictions made using FAHP weights were more accurate than those using AHP weights, regardless of the number of similar cases retrieved.

Table 4

Mean average error (n = 222)

Retrieve (k)	FAHP+CBR	AHP+CBR
1	0.516	0.541
2	0.453	0.458
3	0.412	0.441
4	0.397	0.439
5	0.389	0.435
6	0.380	0.432
7	0.376	0.417
8	0.377	0.424
9	0.375	0.423
10	0.370	0.423
Average	0.404	0.443

5 Conclusion

Web ATM is a convenient internet banking service instead of the physical ATM and a significant payment channel for e-commerce in Taiwan. This study develops the web ATM adoption model which is further used in predicting user intentions to adopt web ATM. There are three factors and 16 criteria in the model. According to the values of FAHP weights from the expert view of point, the factors sorted by priorities concerning web ATM adoption are (1) web ATM security, (2) web ATM knowledge, and (3) web ATM potential value. The six experts consistently agree that security is an essential issue for users to adopt web ATM services. The factor of web ATM security includes transaction data leakage, transaction errors, and computer viruses; whether using the “chip card and password” method to authenticate web ATM security is considered safe; and whether users consider the web ATM on financial transaction websites to be safe.

On the other hand, according to the Eigenvalues of factor analysis from the user view of point, the factors sorted by the explained variance concerning web ATM adoption are (1) web ATM potential value, (2) web ATM knowledge, and (3) web ATM security. This indicates the potential value of web ATM is more contribution to explain users’ intention to adopt web ATM. The factor of web ATM potential value includes whether web ATM save time, provide a variety of services rapidly, offer easier financial management and favorable transaction fees, have higher service capability compared to physical ATM, limit use time, and can provide services from multiple banks simultaneously with a broader range of selectivity.

User knowledge of web ATM is also a decision factor of web ATM adoption, including (1) Users’ awareness of the need to install security components when connecting to the web ATM transaction platform. When accessing the web ATM platform of financial institutions, users are instructed to install security components such as ActiveX, which is a communication component that bridges a remote (bank side) web ATM system with the user’s computer system. (2) Users’ awareness of the installation process of chip card readers (driver installation), because a computer typically cannot recognize a new device, installing drivers is necessary for the chip card reader to establish a link to the computer for communication. (3) Users’ awareness of the hardware specifications required for web ATM use (computer hard disk space, processing speed, memory size, and card reader specifications). (4) Users’ awareness of the browser version specifications required for web ATM use to display content correctly. (5) Users’ awareness of web ATM can run on various computer systems and are highly compatible. The web ATM transaction platform provided by financial institutions in Taiwan supports operating systems such as Windows, Mac, and Linux. If the computer environment is MAC OS X or Linux operating system, users need to install the relevant plugin to execute normally.

AHP is a structured technique method for analyzing multi-attribute decisions. The method dependents on the experts’ point of view. FAHP uses fuzziness to change the values in the scale of expert judgments. Although Saaty and Tran indicates that uncertainty in AHP can be solved by using intermediate values in the 1–9 scale combined with the verbal scale [34], FAHP weights seem to work better in CBR prediction to obtain accurate results than AHP weights in this application. Research such as Wang et al. shows that FAHP and CBR applications are feasible and effective for extracting the style of the product [11]. However, the objective of this research is not to assess the use of the two methods.

According to the values of FAHP weights, the first five important criteria in the three factors are (1) the security concerning “transaction data leakage”, (2) the security concerning “transaction platform safety”, (3) the users’ awareness regarding “authentication process”, (4) the problem concerning “transaction errors” and (5) “computer viruses”. Noteworthy, all concern the security aspect of web ATM, which is consistent with the research indicating the role of perceived risk as a predictor of intention to adopt internet banking [27].

This study developed the web ATM adoption model based on the literature and verified the factor structure of the decision hierarchy from the users’ perspective. This study further proposes a mechanism which integrates CBR with FAHP weights measured by expert pairwise judgments to predict web ATM adoption. The results are helpful for financial institutions to understand the user decision toward web ATM adoption and to further predict the user behavior by referring to expert opinions. This study adopts FAHP to calculate weights for decision criteria, CBR with other weighting methods for the prediction of innovative internet banking services can also be examined in the future.

References

ATM market place, Web ATMs: Is it time to go. https://perma.cc/G3YG-NW6K

Baregheh

, Rowley

and Sambrook

, Towards a multi-disciplinary definition of innovation, Manag Decis 47(8) (2009), 1323–1339.

Bhargava

A.K.

, Fuzzy set theory fuzzy logic and their applications, S. Chand Publishing, 2013.

Brand

M.J.

and Huizingh

E.K.R.E.

, Into the drivers of innovation adoption: What is the impact of the current level of adoption? Eur J Innovat Manag 11(1) (2008), 5–24.

Brasil

L.M.

, de Azevedo

F.M.

, Barreto

J.M.

and Noirhomme-Fraiture

, Complexity and cognitive computing. In International Conference on Industrial, Engineering and Other Applications of Applied Intelligent Systems, 1998, June, 408–417, Springer, Berlin, Heidelberg.

Buckley

J.J.

, Fuzzy hierarchical analysis, Fuzzy Set Syst 17(3) (1985), 233–247.

Chan

S.C.

and Lu

M.T.

, Understanding Internet banking adoption and use behavior: A Hong Kong perspective, J Global Inform Manag 12(3) (2004), 21–43.

Chang

C.W.

, Wu

C.R.

and Lin

C.T.

, Evaluating and controlling silicon wafer slicing quality using fuzzy analytical hierarchy and sensitivity analysis, The Int J Adv Manuf Tech 36 (2008), 322–333.

Deshmukh

and Sunnapwar

, Fuzzy Analytic Hierarchy Process (FAHP) for Green Supplier Selection in Indian Industries, Proceedings of International Conference on Intelligent Manufacturing and Automation, 2019, 679–68.

10.

Durkin

and O’donnell

, Towards a model of adoption in internet banking: Strategic communication challenges, Serv Indust J 25(7) (2005), 861–878.

11.

Wang

, Li

, Dey

, Ashour

A.S.

, Sherratt

R.S.

and Shi

, Case-Based Reasoning for Product Style Construction and Fuzzy Analytic Hierarchy Process Evaluation Modeling Using Consumers Linguistic Variables, IEEE Access 5 (2017), 4900–4912.

12.

Gómez-Vallejo

H.J.

, Uriel-Latorre

, Sande-Meijide

, Villamar?n-Bello

, Pav?n

, Fdez-Riverola

and Glez-Peña

, A case-based reasoning system for aiding detection and classification of nosocomial infections, Decis Support Syst 84 (2016), 104–116.

13.

Hair

J.F.

, Tatham

R.L.

, Anderson

R.E.

and Black

, Multivariate data analysis (Upper Saddle River, NJ: Prentice Hall, 1998)

14.

Hamidi

, A combined fuzzy method for evaluating criteria in enterprise resource planning implementation, In Intelligent Systems: Concepts, Methodologies, Tools, and Applications, 2018, 639–670, IGI Global.

15.

C.T.B.

and Wu

D.D.

, Online banking performance evaluation using data envelopment analysis and principal component analysis, Comput Oper Res 36(6) (2009), 1835–1842.

16.

Hsu

C.I.

, Chiu

C.C.

and Hsu

P.L.

, Predicting information systems outsourcing success using a hierarchical design of case-based reasoning, Expert Syst Appl 26 (2004), 435–441.

17.

IXresearch EC-Write Paper, The overview of usage of E-commerce retail websites, InsightXplorer Biweekly Report 106 (2018), 12–20. https://perma.cc/RUF9-9FNZ

18.

Kahraman

, Cebeci

and Ulukan

, Multi-criteria supplier selection using fuzzy AHP, Logistics Information Management 16(6) (2003), 382–394.

19.

Kaiser

H.F.

, An index of factorial simplicity, Psychometrika 39(1) (1974), 31–36.

20.

Kolodner

, Case-based reasoning (Morgan Kaufmann, 2014).

21.

Kubler

, Robert

, Derigent

, Voisin

and Traon

Y.L.

, A state-of the-art survey & testbed of fuzzy AHP (FAHP) applications, Expert Syst Appl 65 (2016), 398–422.

22.

Leekwijck

W.V.

and Kerre

E.E.

, Defuzzification: criteria and classification, Fuzzy Set Syst 108 (1999), 159–178.

23.

Lin

C.T.

, Chen

C.B.

and Ting

Y.C.

, A Green Purchasing Model by Using ANP and LP Methods, J Test Eval 40(2) (2012), 203–210.

24.

Lin

C.T.

, Lee

and Wu

C.S.

, Fuzzy group decision making in pursuit of a competitive marketing strategy, Int J Inform Tech Decis Making 9(2) (2010), 281–300.

25.

Lin

C.T.

, Wu

C.R.

and Chen

H.C.

, The study of construct key success factors for the Taiwanese hospitals of location selection by using the fuzzy AHP and sensitivity analysis, Int J Inform Manag Sci 19(1) (2008), 175–200.

26.

Lin

Y.H.

and Chen

L.W.

, Fuzzy Approach of Linguistic Variable and Synthetic Decision for IPA on the Multi-Facet Evaluation, International Journal of Kansei Information 1(1) (2010), 59–66.

27.

Martins

, Oliveira

and Popovic

, Understanding the Internet banking adoption: A unified theory of acceptance and use of technology and perceived risk application, Int J Inform Manag 34(1) (2014), 1–13.

28.

MIC, Online Shopping Consumer Survey: 2016 Online Shopping Average NT$27,715. https://perma.cc/GJ7Z-8SP4

29.

Qin

P.P.

, Liu

C.J.

, Shu

J.W.

and Yan

L.Y.

, A case-based reasoning cost estimating model of hydraulic cylinder remanufacturing, Design, Manufacturing and Mechatronics 2017, 232–238.

30.

Reisi

, Afzali

and Aye

, Applications of analytical hierarchy process (AHP) and analytical network process (ANP) for industrial site selections in Isfahan, Iran, Environ Earth Sci 77 (2018), 537.

31.

, Russo

F.S.M.

and Camanho

, Criteria in AHP: a systematic review of literature, Procedia Comput Sci 55 (2015), 1123–1132.

32.

Saaty

L.T.

, Decision making with the analytic hierarchy process, Int J Serv Sci 1(1) (2008), 83–98.

33.

Saaty

R.W.

, The analytic hierarchy process-what it is and how it is used, Math Model 9 (1987), 161–176.

34.

Saaty

T.L.

and Tran

L.T.

, On the invalidity of fuzzifying numerical judgments in the Analytic Hierarchy Process, Math Comput Model 46 (2007), 962–975.

35.

Samuel

O.W.

, Asogbon

G.M.

, Sangaiah

A.K.

, Fang

and Li

, An integrated decision support system based on ANN and Fuzzy_AHP for heart failure risk prediction, Expert Syst Appl 68 (2017), 163–172.

36.

Sukkar

A.A.

and Hasan

, Toward a model for the acceptance of internet banking in developing countries, Inform Tech Dev 11(4) (2005), 381–398.

37.

Syed Mustapha

S.M.F.D.

, Case-based reasoning for identifying knowledge leader within online community, Expert Syst Appl 97 (2018), 244–252.

38.

Taiwan News, Taiwan has highest ATM density in the world, https://perma.cc/7P43-QE5Q

39.

Tapia

J.F.

and Samsatli

, A Fuzzy Analytic Hierarchy Process (FAHP) Approach to Multi-Objective Optimisation of Oil Palm Value Chains, Comput-Aided Chem En 46 (2019), 817–822.

40.

Tavakol

and Dennick

, Making sense of Cronbach’s alpha, Int J Med Educ 2 (2011), 53–55.

41.

Tavana

, Fallahpour

, Caprio

D.D.

and Santos-Arteaga

F.J.

, A hybrid intelligent fuzzy predictive model with simulation for supplier evaluation and selection, Expert Syst Appl 61 (2016), 129–144.

42.

The Telegraph, The story behind the world’s first cashpoint, https://perma.cc/FXY6-LEVF

43.

Wang

Y.S.

, Wu

S.C.

, Lin

H.H.

, Wang

Y.M.

and He

T.R.

, Determinants of user adoption of web ‘Automatic Teller Machines’: an integrated model of ‘Transaction Cost Theory’ and ‘Innovation Diffusion Theory’, Serv Indust J 32(9) (2011), 1505–1525.

44.

Weck

, Klocke

, Schell

and enauver

R?E.

, Evaluating alternative production cycles using the extended fuzzy AHP method, Eur J Oper Res 100(2) (1997), 351–366.

45.

M.C.

, Lo

Y.F.

and Hsu

S.H.

, A fuzzy CBR technique for generating product ideas, Expert Syst Appl 34(1) (2008), 530–540.

46.

Yan

, Yu

and Wang

, Case-based reasoning classifier based on learning pseudo metric retrieval, Expert Syst Appl 89 (2017), 91–98.

47.

Yousafzai

, Pallister

and Foxall

, Multi-dimensional role of trust in Internet banking adoption, Serv Indust J 29(5) (2009), 591–605.

48.

Zadeh

, Fuzzy sets, Information Control 8 (1965), 338–353.

49.

Zeithaml

V.A.

, Parasuraman

and Berry

L.L.

, Problems and strategies in services marketing, J Market 49(2) (1985), 33–46.

50.

Zyoud

S.H.

and Fuchs-Hanusch

, A bibliometric-based survey on AHP and TOPSIS techniques, Expert Syst Appl 78 (2017), 158–181.