Rough set and PSO-based ANFIS approaches to modeling customer satisfaction for affective product design

https://doi.org/10.1016/j.aei.2015.07.005Get rights and content

Highlights

  • Rough set and PSO-based ANFIS approaches are proposed to model customer satisfaction in affective product design.

  • Rough set is used to simplify ANFIS structure and PSO is introduced to enhance the accuracy of modeling.

  • Generated fuzzy rules address fuzziness exists in customer survey data.

  • Generated models show the nonlinear relationships between affective responses and design attributes.

  • The proposed approaches outperform the FLSR, FR, and GP-FR in terms of training errors and validation errors.

Abstract

Facing fierce competition in marketplaces, companies try to determine the optimal settings of design attribute of new products from which the best customer satisfaction can be obtained. To determine the settings, customer satisfaction models relating affective responses of customers to design attributes have to be first developed. Adaptive neuro-fuzzy inference systems (ANFIS) was attempted in previous research and shown to be an effective approach to address the fuzziness of survey data and nonlinearity in modeling customer satisfaction for affective design. However, ANFIS is incapable of modeling the relationships that involve a number of inputs which may cause the failure of the training process of ANFIS and lead to the ‘out of memory’ error. To overcome the limitation, in this paper, rough set (RS) and particle swarm optimization (PSO) based-ANFIS approaches are proposed to model customer satisfaction for affective design and further improve the modeling accuracy. In the approaches, the RS theory is adopted to extract significant design attributes as the inputs of ANFIS and PSO is employed to determine the parameter settings of an ANFIS from which explicit customer satisfaction models with better modeling accuracy can be generated. A case study of affective design of mobile phones is used to illustrate the proposed approaches. The modeling results based on the proposed approaches are compared with those based on ANFIS, fuzzy least-squares regression (FLSR), fuzzy regression (FR), and genetic programming-based fuzzy regression (GP-FR). Results of the training and validation tests show that the proposed approaches perform better than the others in terms of training and validation errors.

Introduction

Affective design has been shown to excite psychological feelings of customers and can help improve the emotional aspects of customer satisfaction. It is an important design strategy to enhance customer satisfaction of new products in customer-driven product development. Design attributes, such as shape and color, evoke the affective responses of customers to products. Products with good affective design can help attract customers and influence their choices and preferences, such as loyalty and joy of use [1], [2]. The process of affective design includes identifying, measuring, analyzing, and understanding the relationship between the affective needs of the customer domain and the perceptual design attributes in the design domain [3]. One of the major processes of affective design is to determine the design attributes settings of new products such that high, or even optimal, customer affective satisfaction of the new products can be obtained. To determine the design attribute settings, customer satisfaction models that relate affective responses of customers to design attributes have to be developed first. However, the modeling process is quite complex as the relationships to be modeled can be highly nonlinear and fuzzy. Modeling customer satisfaction for affective product design has been applied in the industry for various product designs, such as the design of vehicle interior [4], office chairs [5], mobile phones [6], and digital camera [7].

A handful of studies previously attempted to model the relationships between affective responses and design attributes using statistical and artificial intelligence methods. Artificial neural network (ANN) was proposed to model the affective relationship in product design [8], [9]. An interactive evolutionary system based on neural networks was proposed to analyze the aesthetic perceptions of customers and approximate their aesthetic intentions [10]. Chen et al. developed a prototype system for affective design in which Kohonen’s self-organizing map neural network was employed to consolidate the relationships between design attributes and affective dimensions [11]. The main advantage of the ANN is the development of models through learning from data without requiring prior knowledge. Although a trained ANN can possibly provide an accurate prediction or classification, it is known as a ‘black box’ model from which no explicit knowledge of the relationships can be obtained [12].

Multiple linear regression has been used to model affective relationships [13]. The approach is easy to apply, but it assumes that the design attributes in the regression are linear, and the effect of an independent design attribute is the same throughout the entire range of the affective response. A decision support system has been proposed to provide guidelines for optimizing affective satisfaction based on principal component analysis and multiple regression [14]. Petiot and Grognet [15] proposed an explicit modeling method based on a vector field to model affective relationships. You et al. [16] developed the customer satisfaction models for automotive interior material using quantification I analysis. Based on the models, the significance of the design attributes can be identified. Han et al. [17] attempted to evaluate product usability based on statistical regression models that relate usability dimensions and design attributes. However, the above statistical approaches are unable to address the fuzziness involved in the affective responses of customers.

To address the fuzziness of affective modeling, Park and Han proposed a fuzzy rule-based approach to examine customer satisfaction towards office chair designs [18]. They reported that the fuzzy rule-based approach outperformed the multiple linear regression approaches in terms of the number of design attributes to be considered in modeling. A fuzzy expert system with gradient descent optimization was proposed to develop models that relate affective responses to design attributes in fashion product development [19]. Shimizu and Jindo [4] applied a fuzzy regression method to model the relationship between design attributes and affective responses to address the fuzziness of human sensations towards vehicle interior design. Tanaka’s fuzzy regression approach was proposed to model customer satisfaction for improving the design of driver seat [20]. However, the fuzzy regression approach is unable to capture nonlinearity of the modeling. Chan et al. introduced genetic programming into fuzzy regression for modeling affective relationships [6]. An evolutionary algorithm was used to construct branches of a tree representing the structures of a model where the nonlinearity of the model could be addressed and the fuzzy regression was then used to determine the fuzzy coefficients of the model. The limitation of this approach is that the size of the search space increases exponentially with the number of nodes and the tree depth.

The hybrid approaches of fuzzy logic and ANN combine the capability of fuzzy logic in the linguistic representation of knowledge and the adaptive learning capability of ANN for automatic generation and optimization of a fuzzy inference system. Fuzzy neural networks have been introduced to establish the relationships between design attributes and consumer affections [21]. Fuzzy neural networks utilize a series of output nodes of the ANN to emulate a fuzzy membership grade of affection intensity and then determine the aggregate value of customer affection through defuzzification. Hsiao and Tsai [22] proposed a method that enables an automatic product form or product image evaluation by means of a neural network-based fuzzy reasoning and genetic algorithm, which was applied to establish relationships between the design attributes of a new product and the customers’ affective image. An adaptive neuro-fuzzy inference system (ANFIS) was examined by Kwong and Wong [23] to generate explicit customer satisfaction models which can capture the nonlinearity and fuzziness existing in the modeling. Compared with ANN, a set of fuzzy if-then rules with appropriate membership functions and the internal models can be generated based on ANFIS to stipulate input–output pairs explicitly. However, the conventional learning algorithms for ANFIS are gradient descent, in which the calculation of gradients in each step is difficult and the use of chain rules may cause a local minimum. These issues have been shown to affect modeling accuracy. On the other hand, ANFIS is not suitable for the modeling problems that involve a number of inputs. If the number of inputs is large, the number of generated fuzzy rules increases exponentially. These increases would cause long computational time and even execution errors. To overcome the limitation and further improve modeling accuracy of ANFIS, in this paper, rough set (RS) and particle swarm optimization (PSO)-based ANFIS approaches are proposed to modeling customer satisfaction for affective design.

The organization of this paper is as follows: Section 2 describes how the proposed approaches are used to model customer satisfaction for affective design. In Section 3, a case study of mobile phone design is described to illustrate the proposed approaches. The validation of the proposed approaches is shown in Section 4. Finally, conclusions are given in Section 5.

Section snippets

Modeling customer satisfaction using RS and PSO-based ANFIS approaches

To address the deficiency of ANFIS for modeling affective relationships, RS and PSO-based ANFIS approaches are proposed in this research. Since ANFIS is incapable for application in those modeling problems that involve a number of attributes, in the proposed approaches, RS theory is introduced to reduce the number of inputs and determine indispensable design attributes for generating customer satisfaction models. The PSO-based ANFIS approach is introduced to develop nonlinear customer

Case study

A case study of mobile phone design is used in this study to illustrate the proposed approaches to model the relationships between affective responses and design attributes. A total of 32 mobile phones of various brands were selected. Morphological analysis was used to study the representative attributes of mobile phones as numerical data sets. Table 1 shows the nine representative design attributes: top shape, bottom shape, side shape, function button shape, number buttons style, screen size,

Validation of the proposed approaches

A total of 30 validation tests were conducted to further evaluate the effectiveness of the proposed methodology. In each validation test, five data sets were randomly selected from the 32 data sets as the testing data sets, and the remaining 27 data sets were used to develop the customer satisfaction models. The validation tests primarily aim to compare the validation errors of the generated customer satisfaction models based on the proposed approaches with those based on FLSR, FR, and GP-FR.

Conclusion

ANFIS was shown to be an effective approach to generate explicit customer satisfaction models for affective design, and can address both fuzziness and nonlinearity of the modeling. However, it is incapable of modeling the problems that involve a number of inputs. Additionally, the conventional learning algorithm of ANFIS is based on the gradient descent method, which leads to slow convergence of the parameters. In this paper, RS and PSO-based ANFIS approaches to modeling customer satisfaction

Acknowledgement

The work described in this paper was fully supported by a grant from The Hong Kong Polytechnic University (Project No. G-YK81).

References (36)

Cited by (0)

View full text