Customer preference modelling using PLS-regression in mobile phone industry

• Main Author: Chew, Kar Yeong.
• Other Authors: School of Mechanical and Aerospace Engineering
• Format: Final Year Project
• Language: English
• Published: 2010
• Subjects: DRNTU::Engineering::Systems engineering
• Online Access: http://hdl.handle.net/10356/40293
• Institution: Nanyang Technological University

Decision-making has always been a vital part in product development. Especially with the rapid technological advancement today, the competitive market demands designers to make tough choices while the product development cycle is becoming shorter and shorter. In order to capture the market, they have to be able to capture the customers’ wants and needs. Hence, this leads to the development of decision-support systems to aid designers. There are decision-support systems built based decision analysis and multiple-attribute decision making methods. Partial Least Squares regression (PLSR), a recently developed modeling technique, and Customer-revealed Value (CRV) are employed in this study in an effort to develop a more comprehensive decision making method. PLSR is able to construct predictive models using multiple collinear factors while CRV is able to capture customer preference by relating the demand of the product. In this research, the ability of PLSR to be a decision support system is investigated by testing this technique in the mobile phone industry. Predictive PLS models are constructed based on mobile phone attributes as input (X-matrix) and expert and consumer ratings for CRV as responses (Y-matrix). Using VIP, 10 influential attributes are first selected. Then, the PLS models are trimmed down to these 10 attributes before going through validation. The plots of each model are used to validate the PLS models. Then, the prediction ability of the models is examined to further check the validity and reliability of the models. The author proved the effectiveness of using PLSR to model customer preference for the mobile phone industry as the predictive models are valid and able to predict the responses of the validation sets accurately.