Public–Private Partnership Contract Failure Prediction Using Example-Dependent Cost-Sensitive Models

Abstract

Failure of a public–private partnership (PPP) contract could cause heavy losses to sponsors. However, the current machine learning models neglect misclassification costs when predicting PPP contract failure. This research adopts an example-dependent cost-sensitive (ECS) method by customizing the existing algorithms in python libraries. The model treats the opportunity cost and equity loss as the potential cost of misclassifying a successful and failed project, respectively. It is simpler to implement and can identify failed contracts more easily. Profit-oriented and accuracy-oriented metrics, such as cost-savings and F1 score, are used to evaluate the model. A cost-savings of 0.452, representing $863.83 million dollars, is achieved for the test set. This study highlights that the most precise models are not necessarily the most cost-effective. The results can support sponsors in selecting the appropriate models to forecast the outcome of a PPP contract from a financial perspective, contributing to accurate decision-making.