An Enhanced Neuro-fuzzy Approach for Generating Customer Satisfaction Models

Abstract

In this chapter, a new methodology for generating customer requirement models using the approach of neural fuzzy networks is discussed. Non-linear and explicit customer requirement models can be developed using this approach. Unlike standard neural network models, which are black-box in nature, explicit information can be extracted from neural fuzzy network models which are explicit models. The neural fuzzy networks approach is intended to overcome the limitations of the fuzzy regression approach (discussed in Chapter 6 and Chapter 7) which cannot address strong nonlinearity of customer requirements. It can also overcome the limitations of the genetic programming approach (discussed in Chapter 5) which cannot address the fuzzy nature of customer requirements. It consists of a set of fuzzy rules which relate design attributes to customer requirements of new products. Therefore, explicit information can be extracted from rules within the customer satisfaction models, which are generated based on the neural fuzzy network approach. We discuss a rule extraction method for obtaining significant rules to indicate the appropriate ranges of design attributes, in order to achieve reasonable customer requirements in terms of new products. Based on these significant rules, an explicit customer satisfaction model can be constructed. Customer perception of a new product can be understood more easily with the generated customer satisfaction model. An example of a notebook computer design is used to illustrate the methodology. To examine the effectiveness of the proposed methodology, statistical regression was the method against which the results for the new fuzzy approach were benchmarked. Experimental results show that the approach of neural fuzzy networks outperforms statistical regression methods in terms of mean absolute errors and variance of errors. Also, explicit information are more likely to be extracted from the neural fuzzy networks.