A data-driven framework to new product demand prediction: Integrating product differentiation and transfer learning approach

摘要

Predicting the demand for a new product at early stages is crucial in determining successful product designs. However, the lack of market and consumer related data during the early stages make demand prediction incredibly difficult and unreliable, often underestimating or overestimating the product's demand. With increasing global competition and shortening product life-cycle, almost all the new products have some amount of commonality (differentiation) in their design which presents an opportunity to learn from the abundant data available from the predecessor product. In this work, we developed a novel integrated approach for demand prediction, utilizing weighted product differentiation index between the new and the predecessor products and the prior knowledge of the historical demand for the predecessor. The proposed integrated framework employs advanced machine learning algorithms to first model the non-linear and non-stationary relationship between market demand and product differentiation (thus the product design), which we refer as demand differentiation index (DDI) and then utilize this relationship for predicting the initial demand of the new product in early stages. We further propose DDI modified exponential weighted moving average, DDI-EWMA for product life-cycle demand prediction. The efficacy of the model is demonstrated using real data from the automobile industry. (C) 2018 Elsevier Ltd. All rights reserved.