Developing a design support system for the exterior form of running shoes using partial least squares and neural networks

Abstract

The function of sport shoes is to improve sport performance and reduce sports-related injuries. They are commercial products developed by combining sports technology and marketing activities. Numerous studies on research and development, material application, production process improvements, and human physiological measurements for the sport shoe market exist. However, few studies have conducted an in-depth investigation on the design of forms and external appearance for sports shoes.In this study, running shoes are used as the primary research object. In conjunction with Kansei engineering, the correlation between the form of a shoe model and the emotional reaction of consumers is examined in-depth to develop a Kansei prediction system solely for changes in the design of external forms of running shoes. First, the Kansei adjectives for running shoes are obtained and collected using questionnaire methods. Experimental samples of running shoes are developed using the Taguchi experimental design method. Surveys are then conducted using consumer Kansei questionnaires. Dimension reduction is performed on the obtained data using principle component analysis (PCA) and partial least squares (PLS). Finally, neural network (NN), PCA-NN, and PLS-NN are used to establish a predictive model.In this study, we observed that PLS is superior to PCA in performance on dimension reduction because it considers the Y information variable and encompasses causal relationships when converting the original variable X into a latent variable. Data from the PLS regression coefficients shows that the form of the sole is the most significant design factor affecting the emotional response of the consumer regarding the overall outer appearance of the running shoe. Because it is difficult to train neural networks (NNs) when numerous variables exist, the modeling efficiencies of NN with questionnaire data dimension reduction using PCA and PLS are compared. The results indicate that because the elements comprising running shoes correlate with consumer image responses, PCA has a limited ability in reducing dimensions when causal relationships between the variables are disregarded.Therefore, validation results from PCA-NN models are relatively poor. If the correlation between variables is included with PLS, the predictive capability of NNs can be effectively enhanced. The results show that PLS-NN performance in training and validation is superior to PCA-NN. Furthermore, the best predictive result is obtained from the Kansei adjective pair “clumsy-agile.” The prediction accuracy rate of the PLS-NN model improves with increasing reduction performance for PLS. The predictive system developed using PLS-NN maintains stability or robustness for linear data processing and predictive capability with non-linear data. This allows sports shoe designers to provide the best reference combinations for specific product images when creating conceptual designs of novel shoe appearance. The system can present an appropriate external form of running shoe in the user interface as long as an adjective target value is provided by the designer.