Prediction of R&D project outputs incorporating mid-stage signals

초록

As research and development (R&D) projects increasingly require forward-looking performance evaluation, accurately predicting their final outputs has become both a theoretical and practical priority. This study highlights the temporal dynamics of R&D processes, showing that mid-stage outputs such as interim publications are critical indicators of final project performance. Leveraging a national R&D dataset spanning 2010-2021, we build machine learning models incorporating features across project phases to assess the impact of mid-stage signals on final outputs. The empirical results demonstrate that incorporating mid-stage output metrics substantially enhances predictive accuracy, markedly outperforming models that rely solely on data from the proposal or initial stages. Furthermore, to better understand the complex dynamics underlying model performance, we employ explainable artificial intelligence (XAI) techniques to interpret the contribution of individual variables. The analysis shows that funding levels and the size of the research team significantly affect predictive performance, although their relative importance varies across modelling contexts. These findings underscore the importance of conceptualising R&D projects as evolving systems and highlight the methodological and practical value of stage-aware prediction frameworks for improving early evaluation, strategic decision-making, and resource allocation in R&D management.

키워드