Measurement and prediction method of customer requirement importance based on fuzzy DEA-LSSVM model

Abstract

This study addresses the core issue in the field of green product design, the difficulty in accurately determining the importance of customer requirements due to their dynamic and time-varying characteristics. Existing research predominantly employs static weight allocation methods, which face limitations such as neglecting the dynamic evolution of market demands, bias from subjective experience, and challenges in quantifying environmental indicators. To overcome these issues, this paper proposes a method that integrates fuzzy Data Envelopment Analysis (DEA) and Least Squares Support Vector Machine (LSSVM) to determine and predict the final importance of customer requirements in Quality Function Deployment for Environment (QFDE). First, by incorporating the expert experience of the QFD team, the customer preferences are combined with expert experience to correct situations where the importance is zero, determining the basic importance affecting customer requirements; then, based on the fuzzy data envelopment analysis method, the competitive differences of enterprises are evaluated from four aspects: the improvement ratio of customer green demand satisfaction, the feasibility of achieving the improvement goals for customer green demand satisfaction, the selling points of customer green demand, and the ability of enterprises to meet the environmental demand, in order to obtain correction factors for the basic importance, and then correct the basic importance to obtain the final importance of customer requirements; on this basis, the LSSVM model is introduced to predict the corrected final importance in order to grasp the trend of customer requirement changes. Finally, through the case verification, it is found that the final importance changes significantly within the tracking period, only $C_2$ (long battery life) fluctuates less; the LSSVM model can predict the trend of customer requirement importance changes well, with prediction accuracy higher than BP neural networks and SVM models. This study advances the static analytical paradigm of traditional QFD and provides a validated decision-support tool for green design in dynamic market environment.