Performance Analysis of a Stacking Ensemble Machine Learning Model in Forecasting University-industry Collaboration^*

University-industry collaboration has become a pivotal strategy for fostering innovation and economic growth. However, these partnerships are often fraught with high failure rates. Early forecasting of university-industry collaboration performance informs national strategies aimed at improving their success. Recently, standalone machine learning algorithms have been gaining attention for predicting the performance of these collaborations. However, their efficacy in predicting university-industry collaboration performance varies. In this paper, a new approach to forecasting university-industry collaboration using a stacking ensemble model is proposed. Experiments were conducted based on four different machine learning algorithms; random forest, support vector machines, multilayer-perceptron network, and linear regression. These algorithms were used to train the ensemble model for forecasting university-industry performance based on human capital factors. The study utilized a dataset extracted from the global innovation index from 2013 to 2017, containing 639 instances, and 14 attributes. A comparative analysis between the standalone models and the stacking ensemble model revealed that the stacking ensemble model achieved the most promising results with the highest correlation coefficient of 90% and coefficient of determination 81%. The stacking model also obtained the lowest mean absolute error of 4.73 and root mean square error of 6.35. Therefore, we conclude that the proposed approach is superior in predicting university-industry collaboration performance compared to standalone models. The efficacy of stacking ensemble modeling in real-life university-industry collaboration prediction is therefore justified. Drawing upon feature selection, the study also offers valuable insights aimed at improving decision making. Thus, fostering successful university-industry collaborations in many nations.