Please use this identifier to cite or link to this item:
Title: Example-dependent cost-sensitive learning for public- private partnership contract failure prediction
Authors: Wang, Yongqi
Keywords: Engineering::Civil engineering
Issue Date: 2021
Publisher: Nanyang Technological University
Source: Wang, Y. (2021). Example-dependent cost-sensitive learning for public- private partnership contract failure prediction. Doctoral thesis, Nanyang Technological University, Singapore.
Abstract: The public-private partnership (PPP) mode of procurement is increasingly encouraged by governments to deliver public infrastructure and services in developing countries. However, PPP projects have a high risk of encountering a contract failure due to their complex characteristics. Failure of a PPP contract can bring considerable impacts to both the public and private sectors. For example, the public agencies may undergo a heavy financial burden by redeeming the private share, and the private sponsors may receive a bad reputation. Despite various factors determining the final status of a PPP project have been found, some potential determinants, such as countries’ PPP experience and importance in transferring PPP knowledge, have been overlooked. Besides, very few studies have attempted to predict the outcome of a PPP project using project-specific and country-specific factors from a quantitative standpoint. This research is thus carried out to accomplish the following objectives and outcomes. First, a data-driven approach is used to quantify countries’ PPP experience levels through the hybrid Bayesian hierarchical model with uncertainties considered. Detailed data exploration and selection are carried out to clean the data source, followed by the change point detection using the hybrid Bayesian hierarchical model. As a result, different experience levels for one hundred and seventeen developing countries are obtained based on the location of the change points. Four experience levels are suggested for the energy sector, while five levels are found for the transportation and water & sewerage (W&S) sectors. The outcomes can help investors choose the right investment directions. Second, the social network analysis (SNA) technique is used to evaluate the evolution of the PPP knowledge transfer around the world. The importance of both developing and developed countries in transferring the PPP knowledge is represented using the SNA metrics, such as out-degree centrality, in-degree centrality, and the PageRank. The results show that though the importance of a country in transmitting PPP knowledge is changing with time, some are always at key positions. The outcomes provide a guideline on selecting appropriate PPP partners to achieve performance improvement. Finally, to proactively predict the PPP contract failure, an example-dependent cost-sensitive machine learning (ECS-ML) model is proposed in the context of the PPP domain. The conventional machine learning models are first used to obtain the most significant failure factors from a quantitative perspective. Moreover, the ECS-ML model is used to predict PPP contract failure by optimizing the total cost of misclassifications. The opportunity cost and equity loss are treated as the potential cost of misclassifying a successful and failed project, respectively. The performance of the model is evaluated using the profit-oriented and the accuracy-oriented metrics, such as the cost-saving and F1 score. The results highlight that the most precise models are not necessarily the most cost-effective. Besides, the effectiveness of the previously proposed indicators, including countries’ experience levels and importance in transferring PPP knowledge, are also verified.
DOI: 10.32657/10356/154522
Rights: This work is licensed under a Creative Commons Attribution-NonCommercial 4.0 International License (CC BY-NC 4.0).
Fulltext Permission: open
Fulltext Availability: With Fulltext
Appears in Collections:CEE Theses

Page view(s)

Updated on Jun 28, 2022

Download(s) 50

Updated on Jun 28, 2022

Google ScholarTM




Items in DR-NTU are protected by copyright, with all rights reserved, unless otherwise indicated.