Using the fuzzy weighted association rule mining approach to develop a customer satisfaction product form

Abstract

Customer satisfaction is an important indicator of user preferences for products which directly influence customers’ purchase intentions. It is also an essential Kansei factor for enterprises to successfully develop products. Therefore, the objective of this study is to apply the fuzzy weighted association rule mining (FWARM) approach to extract the significant association between customer satisfaction and product form features, thus providing specific parameter guidelines for the enterprise’s business decisions. In previous research, the fuzzy association rule mining (FARM) approach has proved to be a promising way forwards. However, the absence of consideration of the weight of items is always criticized as creating an uninteresting rule with high frequency and low importance. Therefore, in this study the fuzzy Delphi method (FDM) is used to calculate the weight of each item in the early stage of data mining, and filters out items with a high degree of consensus. Then, the FARM method extracts the fuzzy weighted association rule. Taking the household exercise bike as an example, the authors find that handlebar, LCD screen, rack, main outline and pedestal are vital item features. Subsequently, the method is used to identify 14 rules to inform the development of the exercise bike’s form to achieve high customer satisfaction; valuable knowledge support is provided for manufacturers and designers in the initial stage of new product development, potentially improving customer satisfaction and reducing the risk of product failure.

Keywords

Customer satisfaction Kansei engineering Fuzzy Delphi method Fuzzy weighted association rule mining Exercise bike

1 Introduction

In the intensively competitive market environment, new product development (NPD) has moved from ‘production-oriented’ to ‘customer-oriented’. In addition to functionality, price, performance and quality of new products, consumers are increasingly concerned with affective aspects [1]. Good affective features can sharpen the competitive edge of products, and a precise understanding of product affective properties appears particularly important and deserves in-depth investigation [2]. Jordan [3] proposed that the hierarchy of user needs (Fig. 1) is successively: functionality, usability and pleasure from bottom to top and the pleasure experience includes pride, excitement, and satisfaction. Muller [4] believes that customer satisfaction will become a critical factor in the future success of the company, because it directly affects brand loyalty, word of mouth and complaints behaviour. When manufacturers plan to launch new products, they will pay much more attention to the emotional needs of the consumers than on product appearance [5]. Therefore, developing a product form which can satisfy customers’ emotional needs is an essential precondition for highly-reputable and best-selling products. However, in traditional design processes, designers tend to rely on past experience and an aesthetic sense to create products. Decisions which lack customer insight will not meet with great acceptance in the market. Delin et al. [6] pointed out that 80% of newly-launched products each year fail, and only a few products achieve success, so an effective method that can forecast the potential relation between customer satisfaction and the combination of product form features is critical to marketing staff.

Fig. 1

Jordan’s hierarchy of user needs.

In recent years, research concerning Kansei engineering (KE) has been very popular in academia and industry. KE is defined as the technology of systematically transforming customer psychological feelings and images into product design elements based on ergonomics and computer science [7]. It has not only been successfully applied in developing products, but also it has outstanding achievements in patent analysis [8], online shopping websites [9, 10] and other fields. However, their study still has some drawbacks and limitations. For example: (1) the single Kansei adjectives to obtain the optimal design make it difficult to comprehensively summarize customers’ subjectivity and non-linear emotional responses. (2) In previous articles, the relationship between Kansei demands and customer satisfaction is rarely considered. Although there has been some research using two-dimensional models to discuss the influence of Kansei qualities contribution to customer satisfaction [11 –13], this additional process increases the computational time and prolongs the product development cycle. Therefore, in this study the customer’s emotional satisfaction is used as the evaluation index, the evaluation matrix of product form features and customer satisfaction is established and useful decision rules are explored. In addition, many scholars constantly search for advanced mathematical tools to measure the quantitative linking between emotion and design elements. Data mining (DM) is to mine useful knowledge from an enormous database. Knowledge Discovery in Databases (KDD) serves as a classic application in DM, which can mine non-trivial, implicit and potentially useful knowledge. Jiao et al. [14] applied association rule mining (ARM) of DM to the discovery of useful patterns associated with the mapping of affective needs. Shi et al. [15] proposed a type of Kansei knowledge mining method on the basis of ARM, hence identifying Kansei information and a strong association rule set of key mobile phone features. However, ARM can be used for numerical attributes only through the discrete interval method, thus causing a sharp boundary problem. Therefore, Kuok et al. [16] introduced the concept of fuzzy sets and introduced fuzzy ARM (FARM). Consequently, the extractive rule is more natural and easier to understand for people. Using FARM, a number of pieces of research have been done in the stock market [17], the garment industry [18], road traffic [19], electric vehicle [20], and the care industry [21]. However, they only focus on the frequency of items’ attributes so as to calculate the support and confidence, but neglect the relative importance of different items in real applications. Thus, the rules which have high importance and low frequency of occurrence are ignored.

In this paper, we attempt to propose the method of fuzzy weighted association rule mining (FWARM), the composite fuzzy Delphi method (FDM) and FARM to mine the fuzzy weighted rules between customer satisfaction and product design elements to provide concrete parameter guidance for NPD. The main objective is to investigate the consumer’ preferences and develop a product form that satisfies consumer needs. FDM is an anonymous evaluation method for building up an expert groups’ consensus; through experts’ knowledge and experience, the importance of items is evaluated. Moreover, a suitable threshold value is selected to filter the optimal weight item. FARM is used for obtaining the fuzzy association between customer satisfaction and the different weight items. The fuzzy weight association rule mining (FWARM) developed by integrating FDM with FARM, which not only considers the influence of different items weights on fuzzy association rule, but also more accurately obtain the combination of product form elements close to customer satisfaction. The authors take the household exercise bike as an experimental case to demonstrate the prediction model’s validity and applicability.

In the next section, the literature review of related methodologies is outlined. In the third, the framework used in this study is proposed. In the fourth section, the case study is described. In the fifth section, the results and discussion are illustrated and conclusion is drawn in section six.

2 Literature review

2.1 Fuzzy Delphi method

The Delphi method was firstly proposed by an American company, RAND in the late 1950 s. It is a group process used to seek, aggregate, and gain consensus on the opinions of a group of panelists [22]. The Delphi method allows each member to state an intuitive judgment in terms of certain specific issues, hence realizing the function of drawing on the wisdom of the masses. Huang et al. [23] used the Delphi method and KE to select appropriate assessment criteria for trade show design. Miashiewicz and Kozar [24] incorporated expert opinion through the use of Delphi methodology and integrated personas into the design processes for representing and communicating customer needs. Cho and Lee [25] adopted Delphi method and fuzzy analytic hierarchy process (AHP) to develop a new technology product evaluation model for assessing commercialization opportunities. However, this method needs at least two rounds of repetitive surveys, which is not only quite time consuming and costly, but also reduces the experts’ response rate. In addition, the Delphi method selects crisp values in statistics, which are inappropriate for presenting subjective and imprecise experts’ opinions [26]. Therefore, Ishikawa et al. [27] combined fuzzy theory and the Delphi method and applied the concept of cumulative frequency distribution and bi-triangular fuzzy arithmetic to forecast expert’s consensus. Lin [28] applied FDM and the fuzzy AHP to determinate the criteria weight in a fashion design scheme evaluation system. Liu and Chiu [29] adopted FDM to establish the indeces for assessing the culture and creative industries’ positive influence on urban competitiveness and Tseng and Bui [30] integrated a series of measures including FDM to identify the key eco-innovation attributes for enhancing industrial symbiosis performance. Based on qualitative data such as experts’ opinions, EImousalami et al. [31] used FDM to collect and initially rank the cost drivers of the Field Canals Improvement Project. While Tseng et al. [32] utilized FDM and analytical network processes to develop and evaluate the performance of sustainable service supply chain management under uncertainty.

Through reviewing the relevant literature, we find that FDM has the merits of rapidly collecting the required important items and solving the fuzziness problem of experts’ consensus. However, little of the literature is related to the application of FDM in the fitness equipment field, and the attribute weight of the early stage of DM. Therefore, the authors attempted to carry out a first-stage data analysis using this method, evaluating the key items index of the exercise bike’s form features.

2.2 Fuzzy weighted association rule mining

ARM is one of the most important algorithms in DM. Its main purpose is to find the interrelations between data in an enormous database. However, this kind of method has been widely used in many fields [33, 34]. Sangelkar et al. [35] explored the applicability of ARM to efficiently extract rules for improving universal design. Noh et al. [36] used ARM combined with data cube for an in-depth analysis of energy efficiency related factors in commercial buildings. Rekik et al. [37] presented a method based on ARM to study interdependencies between criteria and the categories of a web site. Jiang et al. [38] introduced ARM and multi-objective particle swarm optimisation to generate association rules for supporting affective design. However, traditional ARM adopts the concept of binary values for the partition of item attributes, so there is a problem with giving too much attention to or neglecting the issue of boundary values. Therefore, many scholars propose combining fuzzy theory with ARM to develop a fuzzy association rule to solve the above-mentioned problems [39 –41]. However, regardless of whether ARM or FARM, support and confidence are calculated by counting the frequency of occurrence of items in the transaction. When the value is greater than or equal to the minimum support and confidence, it can be identified as a strong association rule. This kind of method neglects the problem of the different degrees of importance in different items, thus causing the missing of valuable information in real operational situations. Consequently, Cai et al. [42] firstly introduced two new algorithms to solve the problem of mining weighted association rules. Later, Yue et al. [43] proposed a method to mine fuzzy association rules with weighted items. Hong et al. [44] proposed a new mining approach to extract fuzzy linguistic weighted association rules from quantitative transactions. Kaya et al. [45] employed genetic algorithms to optimize membership functions for FWARM. Xue et al. [46] adopted different weights to traditional FARM to improve the accuracy of regional soil quality assessment. Nithya and Duraiswamy [47] combined the gain ratio based weight and the FARM classifier for a medical diagnostic interface. Nithya and Duraiswamy [48] utilized the correlation and gain ratio based average ranking feature selection followed by the FWARM classifier to diagnose the health care data set. Although these FWARM methods have different ways of extracting item’s weights, they have achieved good results in practical operations.

Through the aforementioned work, in order to overcome the defects of assuming uniform weight in previous DM project, in FWARM it is proposed to assign different weights to different item attributes, thus discovering the precise rules of interest. In this paper, the authors apply the FWARM approach to extract the interactive relationship between customer satisfaction and the exercise bike’s form features, in order to provide theoretical guidance for decision makers.

3 Proposed framework

As Fig. 2 shows, a comprehensive research framework which connects FDM and FWARM is proposed; understanding the relationship between customer satisfaction and form features of the target product is achieved. The designer can take the specific form parameters under positive emotional links for reference, facilitating innovative concept development. The process can be divided into two phases as follows:

Fig. 2

Research framework.

(1) Initially, FDM is applied to filter irrelevant or unimportant item attributes; the importance value of key items is set.

(2) Then, FWARM is adopted to extract the fuzzy weighted association rule between customer satisfaction and the specific design elements.

3.1 Fuzzy Delphi method

The fuzzy Delphi method is a concept which is based on the Delphi method incorporating fuzzy theory. It can be used to solve the fuzzy problem of the degree of experts’ consensus, which not only reduces the number of back-and-forth questionnaires, but also improves the efficiency and quality of questionnaires. In this phase, FDM is mainly applied to integrate the opinions of producers and end-users. Then the key item criteria with a high degree of consensus are screened out. Different FDM calculation methods are introduced in many previous studies. In this research, the authors adopt the fuzzy integration Delphi method which is proposed by Ishikawa et al. [27] to evaluate the key items of products. The main steps are as follows:

(1) Establish all the evaluation criteria. Firstly decompose the products into multi-level product features by using morphological analysis. The feature of each product is defined as an evaluation criterion.

(2) Collect data concerning the group decisions by selecting 7-15 experts to assign different interval ranges from 0-10 for the product criteria. The minimum value and maximum value in this interval show respectively the experts’ most optimistic cognition scores and most conservative cognition scores for the quantitative score of the evaluation criteria; the higher value demonstrates that experts think that this criterion is of higher importance.

(3) Establish the triangular fuzzy numbers (TFN) on the basis of experts’ evaluation from step (2). The most optimistic cognition scores can be indicated as O = (L, M, R) by TFN. L, M, R can respectively indicate the minimum value, geometric mean and maximum value in the group. The most conservative cognition scores can be indicated as C = (l, m, r) by TFN. l, m, r respectively indicate the minimum value, geometric mean and maximum value in the group. The double TFN which is composed of those two TFN can then be used to test the consistency of the experts’ opinions, see Fig. 3.

Fig. 3

Double triangular fuzzy numbers.

(4) Consistency test. By comparing the relationship between two TFN, the consistency of the experts’ opinion can be tested. In general, the relationship between these two TFN will have two outcomes: firstly, if the two TFN are not overlapping, namely L ≥ r, which indicates that the interval value of experts opinions have consensus, the opinions of experts are consistent. Therefore, the importance value of evaluation criterion j, 1 ≤ j ≤ m is G_j, which is the arithmetic mean value of M and m_, namely G_j = (M + m)/2. Secondly, if the two TFN are overlapping, namely L ≺ r, and when the interval value of L and r is between m and M, experts’ opinions are consistent and the important value G_j = (M + m)/2. When the interval value of L and r is greater than the interval value of m and M, the experts’ opinions are not consistent. Data collection utilising questionnaires needs to be repeated until the evaluation results reach the experts’ consensus.

(5) Screen out the key evaluation items. Through the above-mentioned FDM, the predicted value G_j of No. j’s evaluation value can be calculated. By means of the threshold value α, the key evaluation items are screened out. The set of threshold value can be determined in accordance with different demands. If G_j ≥ α, the criterion of No. j is the evaluation item. If G_j ≺ α, this criterion can be eliminated. The final weight W_j of key item j can be calculated as: $W_{j} = \frac{G_{j}}{\sum_{j = 1}^{m} G_{j}}$ (1)

3.2 Fuzzy weighted association rule mining

In the second phase of this research, we propose a new fuzzy weighted data-mining algorithm which connects FDM and FARM. The interesting association rules are extracted from the numerical data. The item weights are integrated during the DM process, which makes the result more realistic and effective. In order to be convenient for readers to understand the operational rules, the definitions of some common notations are as follows (Table 1):

Table 1
Definitions of notations in FWARM [44]

Notation Defined

n The total number of data

m The total number of evaluation items

D _i The i-th record in the historical database, 1 ≤ i ≤ n

A _j The j-th item, 1 ≤ j ≤ m

h The number of fuzzy regions for each item

R _jl The l-th fuzzy region of A_j, 1 ≤ l ≤ h

V _ij The quantitative value of A_j in D_i

f _ij The fuzzy set converted from V_ij

f _ijl The membership value of V_ij in region R_jl

count _jl The summation of f_ijl, 1 ≤ i ≤ n

${wsup}_{j}$ The fuzzy weighted support of item A_j

wconf _R The fuzzy weighted confidence of rule R

α The predefined fuzzy weighted set of minimum supports

β The predefined fuzzy weighted set of minimum confidence

W _j The weight for important of item A_j

C _r The set of candidate weighted itemsets with r items

L _r The set of frequent weighted itemsets with r items

Notation	Defined
n	The total number of data
m	The total number of evaluation items
D _i	The i-th record in the historical database, 1 ≤ i ≤ n
A _j	The j-th item, 1 ≤ j ≤ m
h	The number of fuzzy regions for each item
R _jl	The l-th fuzzy region of A_j, 1 ≤ l ≤ h
V _ij	The quantitative value of A_j in D_i
f _ij	The fuzzy set converted from V_ij
f _ijl	The membership value of V_ij in region R_jl
count _jl	The summation of f_ijl, 1 ≤ i ≤ n
${wsup}_{j}$	The fuzzy weighted support of item A_j
wconf _R	The fuzzy weighted confidence of rule R
α	The predefined fuzzy weighted set of minimum supports
β	The predefined fuzzy weighted set of minimum confidence
W _j	The weight for important of item A_j
C _r	The set of candidate weighted itemsets with r items
L _r	The set of frequent weighted itemsets with r items

The algorithm of FWARM can be divided into the following six steps, the detailed elaborations are as follows:

Input: A set of research data about n quantitative values. The date of each line have m items; in accordance with the FDM of last phase, the weight W_j of each item can be generated along with the corresponding membership function of each item; the predefined fuzzy weighted minimum support α and the minimum confidence β.

Output: A set of fuzzy weighted association rules.

Step 1: Through specified fuzzy membership function, convert the original data V_ij of each item A_j in the database into the fuzzy set f_ij’s presentation way: $(\frac{f_{ij 1}}{R_{j 1}} + \frac{f_{ij 2}}{R_{j 2}} + \dots + \frac{f_{ijh}}{R_{jh}})$ (2)h is the number under the region A_j. R_jl represents No. l’s fuzzy region of No. j item; f_ijl represents the fuzzy membership value of V_ij in the fuzzy region R_jl.

Step 2: Count each fuzzy region of each item respectively; the formula is as follows: ${count}_{jl} = \sum_{i = 1}^{n} f_{ijl}$ (3)

Step 3: Multiply numbers of occurrences (fuzzy counts) from step (2) by the corresponding item weight, thus getting the fuzzy weighted support; the formula is as follows: $\underset{jl}{wsup} = W_{j} \times {count}_{jl}$ (4)

Step 4: In accordance with the fuzzy regions of all items, the candidate itemset C₁ will be established. Compare all the attributes of C₁ with the predefined minimum support α. If it is greater than or equal to α, it will be included in frequent 1-itemsets, L₁, which is as follows: $L_{1} = {\underset{jl}{wsup} \geq α, 1 \leq j \leq m}$ (5)

The user needs to input different settings in different phases for the support value α.

Step 5: If L₁ is not null, then move to the next step; otherwise end this algorithm.

Step 6: It is similar to Apriori algorithm [49]. Set r = 1, join frequent itemset L_r, thus creating candidate item set C_r+1. When the two attributes are joined together in the L_r, if the two attributes are from the same item, they can’t be joined together.

Step 7: s represents the (r + 1)-itemset of candidate item set C_r+1, which includes the attributes of s₁, s₂, …, s_r+1, and then operate the following sub-steps:

(a) When the frequent itemset L_r is joined, the intersection operation among fuzzy sets is adopted: $μ_{A \cap B} (x) = min {μ_{A} (x), μ_{B} (x)}, \forall x \in X$ (6)

(b) In the data records D_i of each line, calculate the weighted fuzzy set Wf_is of s: ${Wf}_{is} = {Min}_{j = 1}^{r + 1} W_{j} \otimes {Min}_{j = 1}^{r + 1} f_{{is}_{j}}$ (7)W_j is the weight of No. j item, f_{is
_j} is the membership value of D_i in region s_j.

(c) Calculate the fuzzy weighted support of itemset s ( $\underset{s}{wsup}$ ): $w sup_{s} = \sum_{i = 1}^{n} {Wf}_{is}$ (8)

(d) Check whether the weighted support $\underset{s}{wsup}$ of itemset s is greater than or equal to the fuzzy weighted minimum support α. If it matches the condition, put s into frequent (r + 1)-itemsets L_r+1.

Step 8: If L_r+1 is a null, then execute the next step; otherwise, set r = r + 1 and repeat steps 6 to 8.

Step 9: Construct the association rules from each frequent weighted q-itemset s with items (s₁, s₂, …, s_q), using the following sub steps:

(a) Construct all the possible association rules as follows: $s_{1} \land \dots \land s_{j - 1} \land s_{j + 1} \land \dots \land s_{q} \to s_{j}, j = 1 to q$ (9)

(b) Calculate the fuzzy weighted confidence value wconf_R of each possible association rule R as: ${wconf}_{R} = \frac{{count}_{s}}{{count}_{s - s_{j}}} \otimes W_{s}$ (10) Where ${count}_{s} = \sum_{i = 1}^{n} ({Min}_{k = 1}^{q} f_{{is}_{k}})$ and $W_{s} = {Min}_{i = 1}^{q} W_{j}$

(c) Check whether the fuzzy weighted confidence value wconf_R of association rule R is greater than or equal to fuzzy weighted minimum confidence β. If it meets the conditions, input this interesting rule.

4 Case study

The improvement of living standards increases people’s pursuit for health, which directly promotes the prosperity and development of the fitness equipment market. Taking China as an example, the market size of the fitness equipment industry in 2016 was 32.895 billion Yuan. In contrast with 31.197 billion in 2015, which has increased 5.44% with year-on-year growth [50]. The household upright exercise bike is one of the most popular types of aerobic equipment among sports enthusiasts. It is not influenced by the climate and can simulate the experience of outdoor cycling. However, family members serve as the non-professional users; they don’t have a strong expectations of the function of the bike and therefore, pay more attention to the price and product form style [51]. In most of the previous research concerning exercise bikes only the function level has been considered [52 –54]. The relationship between customer satisfaction and the form features of exercise bikes is rarely discussed in the literature, making this one of the main contributions of this research.

4.1 Stage 1: Establish the importance of key items

4.1.1 Deconstruct the item features of exercise bike

Five graduate students had at least three years’ product design experience with a background in industrial design were recruited to apply the morphological analysis to decompose the target product form. The exercise bike can be divided into approximately six main features, which are respectively Handlebar, LCD Screen, Rack, Main outline, Flywheel, Pedestal. Each feature type can be further divided into three types, which are shown in Fig. 4.

Fig. 4

Form feature analysis of exercise bike.

4.1.2 Evaluate the importance weights of key items

In this phase, invited ten high-involvement experts (five male and five female were invited to participate) including fitness enthusiasts, product managers, etc., and applied the FDM to evaluate the weight of the items which make up household exercise bike. The experts used the interval range from 0-10 to evaluate the above-mentioned six features. In order to analyse the subjective judgement of experts, the double TFN is used to test the consistency of the experts’ opinions. In accordance with the operation steps of FDM, the grey interval of each item’s features were produced and the item’s weight value G_j (shown as Table 2). In order to screen out the key items, the threshold α = 4.5 was set in this research. By comparing the predicted value and threshold value in turn, we find that the predicted value of the Flywheel G_j ≺ 4.5, lead to it being eliminated. Other criterion are all greater than the threshold value α and are therefore kept as the key items. In order to make the weighted result more direct, we apply the normalized method to convert the absolute value G_j into the relative value W_j. Taking the Handlebar as an example, its relative weight value W₁ = 5.79/(5.79 + 6.05 + 5.66 + 7.81 + 4.54) =0.194. It is worth noting that satisfaction serves as an evaluation item, which needs a subjectivity evaluation in order to deduce the subsequent weighted rule. Therefore, the authors set its weight as the mean value of top-five products’ total item weight, which is 0.2. Therefore, the final result of all items are shown as Table 3.

Table 2
Corresponding features of each item in TFN

Items Conservative value Optimistic value Gray interval

(l, m, u) (L, M, U) (L, u) (m, M) G ^j

Handlebar (3.00, 4.41, 6.00) (5.00, 7.16, 9.00) (5, 6) (4.41, 7 . 16) 5.79

LCD Screen (3.00, 4.94, 6.00) (5.00, 7.15, 9.00) (5, 6) (4.94, 7.15) 6.05

Rack (3.00, 4.53, 6.00) (5.00, 6.79, 9.00) (5, 6) (4.53, 6.79) 5.66

Frame (4.00, 6.56, 8.00) (7.00, 9.05, 10.0) (7, 8) (6.56, 9.05) 7.81

Flywheel (1.00, 2.92, 5.00) (3.00, 5.58, 9.00) (3, 5) (2.92, 5.58) 4.25

Pedestal (1.00, 3.20, 5.00) (4.00, 5.87, 8.00) (4, 5) (3.20, 5.87) 4.54

Items	Conservative value	Optimistic value	Gray interval
Handlebar	(3.00, 4.41, 6.00)	(5.00, 7.16, 9.00)	(5, 6)	(4.41, 7 . 16)	5.79
LCD Screen	(3.00, 4.94, 6.00)	(5.00, 7.15, 9.00)	(5, 6)	(4.94, 7.15)	6.05
Rack	(3.00, 4.53, 6.00)	(5.00, 6.79, 9.00)	(5, 6)	(4.53, 6.79)	5.66
Frame	(4.00, 6.56, 8.00)	(7.00, 9.05, 10.0)	(7, 8)	(6.56, 9.05)	7.81
Flywheel	(1.00, 2.92, 5.00)	(3.00, 5.58, 9.00)	(3, 5)	(2.92, 5.58)	4.25
Pedestal	(1.00, 3.20, 5.00)	(4.00, 5.87, 8.00)	(4, 5)	(3.20, 5.87)	4.54

Table 3

Relative weights of key items

Key items	Handlebar	LCD screen	Rack	Frame	Pedestal	Satisfaction
W _j	0.194	0.203	0.190	0.262	0.152	0.200

4.2 Stage 2: Extract the fuzzy weighted association rule between customer satisfaction and product features

4.2.1 Select the representative samples

The authors mainly selected samples of listed products from 2015–2018. Through websites, magazines, books and other channels, 74 samples were preliminarily collected. Because of the large quantity of samples, the testees are likely to be over burdened and become bored. Therefore, bitmap software was adopted to eliminate the brands and other variables, thus generating images as shown in Fig. 5.

Fig. 5

The representative samples.

4.2.2 Establish the evaluation matrix between samples and customer satisfaction

The 12 representative exercise bikes which are screened out from last step with customer satisfaction. Then we use the five-point Likert scale as a semantic measurement tool. Point one shows the lowest satisfaction and point five shows the highest satisfaction. The median value from two points to four points show increasing satisfaction. 130 testees (65 male and 65 female) participated in this questionnaire. Then the average value of customer satisfaction is calculated for the analysis of each piece of data in the follow-up experiment. Taking the first exercise bike as an example, the average customer satisfaction rating is 435.5/130 = 3.35.

4.2.3 Interesting fuzzy weighted association rules are mined by FWARM

FWARM was adopted in the new product development to discuss the potential relationship between customer satisfaction and product design elements. Through the predefined minimum support and confidence, interesting knowledge is mined.

Firstly, the raw data were converted into a fuzzy set. The item sets of the exercise bike is a clear division {C1, C2, C3, C4, C6}, which belong to a special fuzzy expression of 0 or 1. Therefore, extra transitions are not needed. However, customer satisfactions are continuous values that needs to be transferred from the predefined fuzzy membership function (see Fig. 6) into the form of Equation (2). E.g. taking the first record of data as an example, the mean value of satisfaction, 3.35 can be expressed as $(\frac{0.15}{R_{s 1}} + \frac{0.85}{R_{s 2}})$ . All of the values, after conversion are shown in Table 4.

Fig. 6

Membership function of item ‘Satisfaction’.

Table 4

Evaluation matrix between representative samples and customer satisfaction

U	C1			C2			C3			C4			C6			Satisfaction
	1	2	3	1	2	3	1	2	3	1	2	3	1	2	3	1	2
1	0	1	0	1	0	0	0	1	0	1	0	0	1	0	0	0.15	0.85
2	1	0	0	0	1	0	1	0	0	0	1	0	0	0	1	0.64	0.36
3	0	0	1	0	1	0	0	1	0	1	0	0	0	1	0	0.00	1.00
4	0	0	1	0	1	0	0	1	0	0	1	0	0	1	0	0.34	0.66
5	0	1	0	0	0	1	0	0	1	0	0	1	0	0	1	0.82	0.18
6	1	0	0	0	1	0	0	1	0	1	0	0	0	0	1	0.89	0.11
7	1	0	0	1	0	0	0	0	1	0	0	1	0	0	1	0.11	0.89
8	0	0	1	0	0	1	0	0	1	0	1	0	1	0	0	0.13	0.87
9	1	0	0	0	1	0	0	1	0	0	0	1	1	0	0	0.29	0.71
10	0	1	0	0	1	0	0	0	1	0	1	0	1	0	0	0.00	1.00
11	1	0	0	1	0	0	1	0	0	1	0	0	0	1	0	1.00	0.00
12	0	0	1	1	0	0	1	0	0	0	1	0	1	0	0	0.70	0.30

Secondly, Equations (3–4) is adopted to sum up the fuzzy interval of each item and then multiply by the corresponding weight, thus generating the fuzzy weighted support. For example, the fuzzy weighted support of C11 is $\underset{11}{wsup} = 0.194$ ⊗ ({1 + 1 +1 + 1 +1}) =0.97. Similarly, the other fuzzy intervals are worked out and compiled as the candidate 1-itemset, C1. As Table 5 shows, the minimum support of each phase is different, so users set different threshold values of α (Table 6). The result shows that the fuzzy weighted support of grey zones in Table 5 are all greater than the minimum support. Therefore they are categorized in frequent 1-itemsets, L₁.

Table 5

Candidate 1-itemsets α = 0.7

Items	Fuzzy weighted support
C11	0.970
C12	0.582
C13	0.776
C21	0.812
C22	1.218
C23	0.406
C31	0.570
C32	0.950
C33	0.760
C41	1.048
C42	1.310
C43	0.786
C61	0.760
C62	0.456
C63	0.608
S11	1.014
S12	1.386

Notes: Grey zones are the frequency itemsets which are greater than minimum support.

Table 6

The threshold support for each period of frequency itemsets

Predefined support count	L ₁	L ₂	L ₃	L ₄
α	0.7	0.6	0.4	0.3

L₁ is not a null, so apriori theory is adopted to join any two items in L₁, thus totally generating candidate 2-itemsets. According to Equations (6–8), their fuzzy weighted supports are re-evaluated. Altogether 11 frequency 2-itemsets are generated (Table 7). In a similar way, formulate a higher level of itemset until there are no available combinations to be found. Tables 8 and 9 show the frequency 3-itemsets and 4-itemsets. Therefore in total 17 frequency itemsets are eligible for this case study.

Table 7

Frequent 2-itemsets α = 0.6

Items	Fuzzy weighted support	Items	Fuzzy weighted support
C11, S11	0.782	C33, S12	0.660
C11, S12	0.623	C42, S11	0.759
C13, S12	0.732	C42, S12	1.116
C22, S11	0.721	C43, S12	0.614
C22, S12	1.047	C61, S12	0.758
C32, S12	0.815

Table 8

Frequent 3-itemsets α = 0.4

Items	Fuzzy weighted support	Items	Fuzzy weighted support
C11, C22, S11	0.533	C22, C42, S12	0.540
C13, C42, S12	0.538	C42, C61, S12	0.454
C22, C32, S12	0.625

Table 9

Frequent 4-itemsets α = 0.3

Items	Fuzzy weighted support
C13, C42, C61, S12	0.302

Finally, rule filtering was implemented to ensure the consistency of the rule structure. The product item features form the antecedent part of the rules and customer satisfaction forms the consequent part. Therefore in accordance with Equations (9–10), each possible fuzzy weighted association rule is generated and their fuzzy weighted confidence wconf_R is worked out. For example, the confidence of rule if C11, then S11 can be calculated as follows: $\frac{C 11 \cap S 11}{C 11} \otimes W_{1} = \frac{2.93}{5} \otimes 0.194 = 0.114$

Similarly, the confidence of all rules can be obtained as Table 10 shows. In this period, the minimum threshold confidence is predefined β = 0.085. After eliminating rules which are below the threshold value, 14 interesting rules are extracted (as depicted as grey zones in Table 10), which provides some references about decision knowledge for enterprises and designers at the initial stage of product development. Only when the result (with strong support and high confidence) is greater than the threshold value can it be considered as an effective rule, so the performance of each retention rule is guaranteed.

Table 10

All the possible generated fuzzy weighted association rules β = 0.085

Rules	Confidence
if C11, then S11	0.114
if C11, then S12	0.080
if C13, then S12	0.137
if C22, then S11	0.072
if C22, then S12	0.128
if C32, then S12	0.127
if C33, then S12	0.140
if C42, then S11	0.072
if C42, then S12	0.128
if C43, then S12	0.119
if C61, then S12	0.113
if C11, C22, then S11	0.118
if C13, C42, then S12	0.118
if C22, C32, then S12	0.118
if C22, C42, then S12	0.135
if C42, C61, then S12	0.110
if C13, C42, C61, then S12	0.089

Note: Grey zones are the strong association rules which satisfy the minimum confidence value.

5 Discussion

In the extremely competitive, market companies hope that their newly-listed products attract market attention leading to enhanced sales. However, customers desire products which satisfy their expectations. Therefore, consumer satisfaction with a product is an important evaluation indicator, which can lead to repeated purchases and favourable word-of-mouth publicity. Previous KE studies have applied individual affective factors (i.e., beautiful, fashionable) to establish a mapping relationship with product variables. Nevertheless, the strength of a single emotional need is not necessarily related to user satisfaction, and it is difficult to directly influence the buying behaviors of customers. Hence, we considered all the other positive emotional perceptions as a whole evaluation index, namely consumer satisfaction.

Currently, The traditional ARM with binary (boolean) attributes are lack of sharp boundary value conversion process. Since ARM is not suitable for classification purpose, FARM adopted a level-wise technology to find fuzzy association rules from a Kansei database. But in real-world applications, each item usually has a weight; FARM does not consider the different importance between items, so it is easy to get the rules with high frequency but boring. The FWARM proposed in this paper considers the ambiguity of binary value data and the weight of each item to reflect their significance to the user. The ‘If-Then’ result is more accurate and meets the real needs of customers. FWARM has been proposed by many scholars to solve the problems in transactions, census database, healthcare data, soil quality assessment and other research areas. Like Shamgasundaram and Nithya [55] using FWARM to analyzed with benchmark data collected from medical data. Altuntas et al. [56] proposed FWARM based solution approaches to facility layout problem in cellular manufacturing system. Methods reported in these articles use different methods to extract the item weight. In the simplest, i.e., managers directly evaluate the items’ importance [44], which is too dependent upon the subjective evaluation of experts. Other genetic algorithm based clustering methods [45], correlated gain ratio [48] and global weights [46] are computationally more complicated. Therefore, for the first time, this paper innovatively explores the FDM method in multi-criteria decision making in the early stage of DM to accurately obtain the attribute weights of different items by using a group-decision technique.

In order to consider both the above issues simultaneously, this paper proposes a FWARM approach that consists of FDM and FARM to discover hidden patterns between customer satisfaction and product feature combinations, thus provide concrete parameter guidelines for enterprises’ business decision. This decision-support model can not only greatly reduce the risks of NPD, but also can assist designers in developing new products which can satisfy customers’ desires. In this research, the exercise bike was taken as an example to verify the efficiency and effectiveness of this systematic method. Fourteen interesting rules which can improve customer satisfaction are found in the results. Twelve of them significantly improve high satisfaction, and two are rules that improve low satisfaction (See Table 11). Designers can use them, in accordance with the optimal parameters to produce the next generation exercise bike. Taking the longest rule as an example, if Handlebar is type 3, Main outline is type 2, Pedestal is type 1, then customers will feel very satisfied. A comparison between this study and previous studies is shown in Table 12. However, several research limitations still exist:

DM is only for deeply mining users’ past preferences, so it is very difficult to create business opportunities with a forward-thinking design.

FDM is subjective weight calculation method. It is easily disturbed by domain experts’ knowledge that causes inaccurate results in the evaluation stage. In the future, information entropy or grey relationship matrix analysis and other objective weight techniques are suggested for use.

By using morphological analysis, many items’ features can be decomposed. In this context, only six items are used to briefly illustrate the feasibility of this method. More specific details can be taken into consideration in the future stage of NPD, such as human-computer interaction (user experience), layout of screen buttons and style of seats.

Table 11
Fourteen rules of interest

Rules Confidence Satisfaction degree

if C11, then S11 0.114 Low satisfaction

if C11, C22, then S11 0.118

if C13, then S12 0.137 High satisfaction

if C22, then S12 0.128

if C32, then S12 0.127

if C33, then S12 0.140

if C42, then S12 0.128

if C43, then S12 0.119

if C61, then S12 0.113

if C13, C42, then S12 0.118

if C22, C32, then S12 0.118

if C22, C42, then S12 0.135

if C42, C61, then S12 0.110

if C13, C42, C61, then S12 0.089

Rules	Confidence	Satisfaction degree
if C11, then S11	0.114	Low satisfaction
if C11, C22, then S11	0.118
if C13, then S12	0.137	High satisfaction
if C22, then S12	0.128
if C32, then S12	0.127
if C33, then S12	0.140
if C42, then S12	0.128
if C43, then S12	0.119
if C61, then S12	0.113
if C13, C42, then S12	0.118
if C22, C32, then S12	0.118
if C22, C42, then S12	0.135
if C42, C61, then S12	0.110
if C13, C42, C61, then S12	0.089

Table 12

Comparison of this study with existing work

Publication	Market segmentation	Emotional evaluation	Functional evaluation	Causalities between CRs and DEs	Techniques used
This paper		√		√	FDM, FWARM
Ho et al. [17]			√		FARM
Wong et al. [21]			√		LOTs, FARM
Lee et al. [19]			√	√	FARM, RFID
Lee et al. [57]		√		√	DM, FARM
Wang [58]	√		√		TRIZ, ARM, CA
Wang and Nien [59]	√		√	√	MCA, ARM, KNN
Lau et al. [60]			√	√	IPM, FARM
Sowan et al. [18]			√		FCM, FARM
Liao et al. [61]		√		√	DM, ARM
Lam et al. [62]		√		√	ARM, NN
Jiao and Zhang [63]		√		√	DM, ARM
Jiao et al. [14]		√		√	KE, ARM, CA
Huang et al. [23]		√		√	KE, Delphi method
Hsueh and Kuo [64]			√	√	Delphi method, AHP
Zuo et al. [65]			√		LINMAP

Notes: Table describes the differences and connections between this paper and previous papers, especially in emotional or functional evaluations, as well as the use of advanced methods. Notes: CRs: Customer requirements; DEs: Design elements; LoTs: Internet of things; TRIZ: The theory of inventive problem solving; CA: Conjoint analysis; MCA: Multiple correspondence analysis; KNN: K nearest neighbor; RFID: Radio frequency identification technology; IPM: Iterative process mining; FCM: Fuzzy c-mean; NN: Neural network; LINMAP: Linear programming technique for multidimensional analysis of preference.

6 Conclusion

Constantly innovating products and meeting customer preferences are key factors for a company to survive. This study takes consumer satisfaction as a measure indicator, applied the concept of FWARM, replaced the single affective attribute evaluation of KE, systematically and comprehensively explores the correlation between consumer satisfaction and product form features. For the pilot study, we firstly use the FDM method to distinguish important product types and solve the multi-parameter trade-off. Then, we can get the consumer satisfaction product form through the use of FARM. This model can provide designers to think about product design from the standpoint of ‘buyer-dominate’. The research results show that the proposed design model can indeed assist business operators in the quantitative analysis of the decision-making decision, and can maximize enhance consumer satisfaction. The main contributions of this paper can be summarized as follows:

Using consumer satisfaction to cover all perceptual dimensions for NPD, comprehensively sums up the multidimensional nature of consumer’s inner feelings.

The importance degree of key item attributes is confirmed by FDM and irrelevant items are filtered. With this assessment model, it is possible to trade off whether consumers like or dislike the styling features of this product and avoid paying more attention to the non-essential features.

The FWARM approach is firstly used to extract the interaction between customer satisfaction and product form features, to facilitate the designer to correctly identifying design parameters to effectively improve consumer satisfaction.

The experimental result is expressed via fuzzy rules in a more understandable way, and screens out the appeal features of exercise bikes and the combination of form elements, which is very attractive to companies, marketing staff, product planners and other relevant practitioners.

In summary, a novel framework fused FARM and FDM to explore consumer needs and preferences. A case study is used to guide companies in enhancing consumer satisfaction with new products. Since the previous Kansei survey data was collected on a relatively small scale on only one occasion. In the future, extensive and dynamic methods can be used to collect consumer instant evaluations, for example based on online customer reviews.

References

Chan

K.Y.

, Wong

T.C.

and Kwong

C.K.

, Special issue on affective design using big data, Journal of Engineering Design 29 (2018), 353–357.

Chang

and Lee

, A product affective properties identification approach based on web mining in a crowdsourcing environment, Journal of Engineering Design 29 (2018), 449–483.

Jordan

P.W.

, Designing pleasurable products: an introduction to the newhuman factors, NewYork: Taylor and Francis (2000).

Muller

, Gaining competitive advantage through customer satisfaction, European Management Journal 9 (1991), 201–211.

Guo

, Liu

W.L.

, Liu

F.T.

, Wang

and Wang

T.B.

, Emotional design method of product presented in multi-dimensional variables based on Kansei Engineering, Journal of Engineering Design 25 (2014), 194–212.

Delin

, Sharoff

, Lillford

and Barnes

, Linguistic support for concept selection decisions, and Manufacturing 21 (2007), 123–135.

Nagamachi

, Kansei engineering: A new ergonomic consumer-oriented technology for product development, International Journal of Industrial Ergonomics 15 (1995), 3–11.

Wodehouse

, Vasantha

, Corney

, Jagadeesan

and MacLachlan

, Realising the affective potential of patents: A new model of database interpretation for user-centred design, Journal of Engineering Design 29 (2018), 484–511.

, Tian

Z.G.

, Wang

J.W.

, Wang

W.M.

and Huang

G.Q.

, Dynamic mapping of design elements and affective responses: A machine learning based method for affective design, Journal of Engineering Design 29 (2018), 358–380.

10.

Wang

W.M.

, Li

, Liu

, Tian

Z.G.

and Tsui

, Mining of affective responses and affective intentions of products from unstructured text, Journal of Engineering Design 29 (2018), 404–429.

11.

Chen

C.C.

and Chuang

M.C.

, Integrating the Kano model into a robust design approach to enhance customer satisfaction with product design, International Journal of Production Economics 114 (2008), 667–681.

12.

, Jiao

R.J.

, Yang

and Helander

, An analytical kano model for customer need analysis, Design Studies 30 (2009), 87–110.

13.

Kang

, Yang

, Wu

and Ni

, Integrating evaluation grid method and fuzzy quality function deployment to new product development, Mathematical Problems in Engineering (2018), 1–15.

14.

Jiao

, Zhang

and Helander

M.A.

, Kansei mining system for affective design, Applications 30 (2006), 658–673.

15.

Shi

, Sun

and Xu

, Employing rough sets and association rule mining in Kansei knowledge extraction, Information Sciences 196 (2012), 118–128.

16.

Kuok

C.M.

, Fu

and Wong

M.H.

, Mining fuzzy association rules in databases, Acm Sigmod Record 27 (1998), 41–46.

17.

G.T.S.

, Ip

W.H.

, Wu

C.H.

and Tse

Y.K.

, Using a fuzzy association rule mining approach to identify the financial data association, Expert Systems with Applications 39 (2012), 9054–9063.

18.

Sowan

, Dahal

, Hossain

M.A.

, Zhang

and Spencer

, Fuzzy association rule mining approaches for enhancing prediction performance, Expert Systems with Applications 40 (2013), 6928–6937.

19.

Lee

C.K.H.

, Ho

G.T.S.

, Choy

K.L.

and Pang

G.K.H.

, A RFID-based recursive process mining system for quality assurance in the garment industry, International Journal of Production Research 52 (2014), 4216–4238.

20.

Kim

, Han

, Lee

and Park

, Futuristic data-driven scenario building: Incorporating text mining and fuzzy association rule mining into fuzzy cognitive map, Expert Systems with Applications 57 (2016), 311–323.

21.

Wong

, Ho

G.T.S.

and Tsui

, Development of an intelligent e-healthcare system for the domestic care industry, Industrial Management and Data Systems 117 (2017), 1426–1445.

22.

Schmidt

R.C.

, Managing Delphi surveys using nonparametric statistical techniques, Decision Sciences 28 (1997), 763–774.

23.

Huang

M.S.

, Tsai

H.C.

and Huang

T.H.

, Applying Kansei engineering to industrial machinery trade show booth design, International Journal of Industrial Ergonomics 41 (2011), 72–78.

24.

Miaskiewicz

and Kozar

K.A.

, Personas and user-centered design: How can personas benefit product design processes? Design Studies 32 (2011), 417–430.

25.

Cho

and Lee

, Development of a new technology product evaluation model for assessing commercialization opportunities using Delphi method and fuzzy AHP approach, Expert Systems with Applications 40 (2013), 5314–5330.

26.

, Zhang

and Yan

, New doctors ranking system based on VIKOR method, International Transactions in Operational Research 27(2) (2020), 1236–1261.

27.

Ishikawa

, Amagasa

and Shiga

, G.Tomizawa, R. Tatsuta and H. Mieno, The Max-Min Delphi method and fuzzy Delphi method via fuzzy integration, Fuzzy Sets and Systems 55(3) (1993), 241–253.

28.

Lin

, Application of fuzzy Delphi method (FDM) and fuzzy analytic hierarchy process (FAHP) to criteria weights for fashion design scheme evaluation, International Journal of Clothing Science and Technology 25 (2013), 171–183.

29.

Liu

Y.Y.

and Chiu

Y.H.

, Evaluation of the policy of the creative industry for urban development, Sustainability 9 (2017), 1009.

30.

Tseng

M.L.

and Bui

T.D.

, Identifying eco-innovation in industrial symbiosis under linguistic preferences: A novel hierarchical approach, Journal of Cleaner Production 140 (2017), 1376–1389.

31.

Elmousalami

H.H.

, Elyamany

A.H.

and Ibrahim

A.H.

, Evaluation of cost drivers for field canals improvement projects, Water Resources Management 32 (2018), 53–65.

32.

Tseng

M.L.

, Ming

K.L.

, Wong

W.P.

, Chen

Y.C.

and Zhan

, A framework for evaluating the performance of sustainable service supply chain management under uncertainty, Economics 195 (2018), 359–372.

33.

Viktoratos

, Tsadiras

and Bassiliades

, Combining community-based knowledge with association rule mining to alleviate the cold start problem in context-aware recommender systems, Expert Systems with Applications 101 (2018), 78–90.

34.

Aaldering

L.J.

, Leker

and Song

C.H.

, Analyzing the impact of industry sectors on the composition of business ecosystem: a combined approach using ARM and DEMATEL, Expert Systems with Applications 100 (2018), 17–29.

35.

Sangelkar

, Cowen

and McAdams

, User activity-product function association based design rules for universal products, Design Studies 33 (2012), 85–110.

36.

Noh

, Son

, Park

and Chang

, In-depth analysis of energy efficiency related factors in commercial buildings using data cube and association rule mining, Sustainability 9(11) (2017), 2119.

37.

Rekik

, Kallel

, Casillas

and Alimi

A.M.

, Assessing web sites quality: A systematic literature review by text and association rules mining, International Journal of Information Management 38(1) (2018), 201–216.

38.

Jiang

, Kwong

C.K.

, Park

W.Y.

and Yu

K.M.

, A multi-objective PSO approach of mining association rules for affective design based on online customer reviews, Journal of Engineering Design 29 (2018), 381–403.

39.

Rajeswari

A.M.

and Deisy

, Fuzzy logic based associative classifier for slow learners prediction, Journal of Intelligent & Fuzzy Systems 36 (2019), 2691–2704.

40.

Leung

K.H.

, Luk

C.C.

, Choy

K.L.

, Lam

H.Y.

and Carman

K.M. Lee

, A B2B flexible pricing decision support system for managing the request for quotation process under e-commerce business environment, International Journal of Production Research 57(20) (2019), 6528–6551.

41.

Bouaita

, Moussaoui

and Bachari

N.E.I.

, Rainfall estimation from MSG images using fuzzy association rules, Journal of Intelligent & Fuzzy Systems 37 (2019), 1357–1369.

42.

Cai

C.H.

, Fu

A.W.C.

, Cheng

C.H.

and Kwong

W.W.

, Mining association rules with weighted items, The International Database Engineering and Applications Symposium (1998), 68–77.

43.

Yue

J.S.

, Tsang

, Yeung

and Shi

, Mining fuzzy association rules with weighted items, The IEEE InternationalConference on Systems, Man and Cybernetics (2000), 1906–1911.

44.

Hong

T.P.

, Chiang

M.J.

and Wang

S.L.

, Fuzzy weighted data mining from quantitative transactions with linguistic minimum supports and confidences, International Journal of Fuzzy Systems 8 (2006), 173–182.

45.

Kaya

and Alhajj

, Utilizing genetic algorithms to optimize membership functions for fuzzy weighted association rules mining, Applied Intelligence 24 (2006), 7–15.

46.

Xue

Y.J.

, Liu

S.G.

, Hu

Y.M.

and Yang

J.F.

, Soil quality assessment using weighted fuzzy association rules, edosphere 20 (2010), 334–341.

47.

Nithya

N.S.

and Duraiswamy

, Gain ratio based fuzzy weighted association rule mining classifier for medical diagnostic interface, adhana-Academy Proceedings in Engineering Sciences 39 (2014), 39–52.

48.

Nithya

N.S.

and Duraiswamy

, Correlated gain ratio based fuzzy weighted association rule mining classifier for diagnosis health care data, Journal of Intelligent & Fuzzy Systems 29 (2015), 1453–1464.

49.

Agrawal

and Srikant

, Fast algorithm for mining association rules, The international Conference On Very Large Databases (1994), 487–499.

50.

National Bureau of Statistic of the China, Analysis and research on the prediction of scale and prospect of Chinese fitness equipment market, http://www.stats.gov.cn/&num; (accessed August 2016).

51.

Hsiao

S.W.

and Ko

Y.C.

, A study on bicycle appearance preference by using FCM and FAHP, International Journal of Industrial Ergonomics 43 (2013), 264–273.

52.

Mohammadi-Abdar

, Ridgel

A.L.

, Discenzo

F.M.

, Phillip

R.S.

, Walter

B.L.

and Loparo

K.A.

, Test and validation of a smart exercise bike for motor rehabilitation in individuals with parkinson’s disease, Neural Systems and Rehabilitation Engineering 24(11) (2016), 1254–1264.

53.

Scanlon

J.E.M.

, Sieben

A.J.

, Holyk

K.R.

and Mathewson

K.E.

, Your brain on bikes: P3, MMN/N2b, and baseline noise while pedaling a stationary bike, Psychophysiology 54 (2017), 927–937.

54.

Huang

S.Y.

, Yu

J.P.

, Wang

Y.K.

and Liu

J.W.

, Designing an exergaming system for exercise bikes using kinect sensors and Google Earth, Multimedia Tools and Applications 76(10) (2017), 12281–12314.

55.

Shamugasundaram

and Nithya

N.S.

, An Analysis of fuzzy weighted association rule mining from medical data, Asian Journal of Research in Social Sciences and Humanities 6(7) (2016), 471–488.

56.

Altuntas

, Dereli

and Selim

, Fuzzy weighted association rule based solution approaches to facility layout problem in cellular manufacturing system, International Journal of Industrial and Systems Engineering 15(3) (2013), 253–271.

57.

Lee

C.K.H.

, Tse

Y.K.

and Ho

G.T.S.

, Fuzzy association rule mining for fashion product development, Industrial Management and Data Systems 115 (2015), 383–399.

58.

Wang

C.H.

, Using the theory of inventive problem solving to brainstorm innovative ideas for assessing varieties of phone-cameras, Computers and Industrial Engineering 85 (2015), 227–234.

59.

Wang

C.H.

and Nien

S.H.

, Combining multiple correspondence analysis with association rule mining to conduct user-driven product design of wearable devices, Computer Standards and Interfaces 45 (2016), 37–44.

60.

Lau

H.C.W.

, Ho

G.T.S.

, Chu

K.F.

, Ho

and Lee

C.K.M.

, Development of an intelligent quality management system using fuzzy association rules, Expert Systems with Applications 36(2) (2009), 1801–1815.

61.

Liao

S.H.

, Chen

Y.J.

and Deng

M.Y.

, Mining customer knowledge for tourism new product development and customer relationship management, Expert Systems with Applications 37 (2010), 4212–4223.

62.

Lam

H.Y.

, Ho

G.T.S.

, Wu

C.H.

and Choy

K.L.

, Customer relationship mining system for effective strategies formulation, Industrial Management and Data Systems 114(5) (2014), 711–733.

63.

Jiao

and Zhang

, Product portfolio identification based on association rule mining, Computer-Aided Design 37 (2005), 149–172.

64.

Hsueh

S.L.

and Kuo

C.H.

, Factors in an interdisciplinary curriculum for the students of industrial design designing multifunctional products, Science and Technology Education 12 (2016), 1075–1089.

65.

Zuo

W.-J.

, Li

D.-F.

, Yu

G.-F.

and Zhang

L.-P.

, A large group decision-making method and its application to the evaluation of property perceived service quality, Journal of Intelligent & Fuzzy Systems 37(1) (2019), 1513–1527.