Data-driven prediction of contract failure of public-private partnership projects

Abstract

The public-private partnership (PPP) has been adopted by many governments in developing countries to provide better public services. However, PPP projects have a high risk of contract failure. To proactively predict PPP contract failure and obtain the most significant failure factors from a quantitative perspective, this research compared the performance of different combinations of machine learning models and data-balancing techniques. Forty-three project-specific and country-specific factors were examined, and the top 15 were chosen for the transportation, water and sewer, and energy sectors. The results show that the selected model can forecast contract failure with a recall of 75.9%, 73.3%, and 76.2%, respectively. This study showed the effectiveness and applicability of machine learning in predicting PPP contract failure. The results can facilitate decision making by forecasting the probability of PPP contract failure in the early planning stage.