Designing and analyzing a deep-learning-based multi-input model to predict the outcome of R&D projects

Abstract

As in many decision problems, data-driven selection of research and development (R&D) projects is drawing keen interests in the field of R&D project management. One of the central problems in such data-driven approaches is to accurately assess the prospect of a proposed proposal

which projects will succeed? This study presents a deep-learning-based model to predict the outcome of project proposals. We develop multi-input deep-learning models to consider both structured and unstructured data on project proposals and evaluate their effectiveness on outcome prediction. We apply our model to the national R&D project data from the Korean government. In particular, we use categorical and numeric data as well as text descriptions from project proposals to predict whether a funded project materializes into commercialization. The performance of model prediction is presented and the development possibility of the model is discussed.