Predicting funded research project performance based on machine learning

Increasing investment and interest in research and development (R&D) requires an efficient management system for achieving better research project outputs. In tandem with this trend, there is a growing need to develop a method for predicting research project outputs. Motivated by this, using information gathered in the early stage of projects, this study addresses the problem of predicting research projects’ output, which is binary coded as either successful or not. To build the prediction model, we apply six machine learning algorithms: five are well-known supervised learning algorithms and the other is autoML, characterized by its ability to produce a learning model appropriate to the data characteristics on its own, with minimal user intervention. Our empirical analysis with real R&D data provided by the South Korean government over 5 years (2014–8) confirms that the autoML-based model performs better than models based on other machine learning algorithms for this task. We also find that project duration and research funding are important factors in predicting R&D project outputs. Based on the results, our study provides insightful implications leading to a paradigm shift for data-based R&D project management.