مقایسه مدل های طبقه بندی با برچسب چندگانه برای پیش بینی حل اختلاف پروژه

Early forecasting of project dispute resolutions (PDRs) provides decision-support information for resolving potential procurement problems before a dispute occurs. This study compares the performances of classification and ensemble models for predicting dispute handling methods in public–private partnership (PPP) projects. Model analyses use machine learners (i.e., Support Vector Machines (SVMs), Artificial Neural Networks (ANNs), and Tree-augmented Naïve (TAN) Bayesian), classification and regression-based techniques (i.e., Classification and Regression Tree (CART), Quick, Unbiased and Efficient Statistical Tree (QUEST), Exhaustive Chi-squared Automatic Interaction Detection (Exhaustive CHAID), and C5.0), and combinations of these techniques that performed best for a set of PPP data. Analytical results exhibit that the combined technique of QUEST + CHAID + C5.0 has the best classification accuracy at 84.65% in predicting dispute resolution outcomes (i.e., mediation, arbitration, litigation, negotiation, administrative appeals or no dispute occurred). Moreover, as the dispute category and phase in which the dispute occurs are known during project execution, the best classification model is the CART model, with an accuracy of 69.05%. This study demonstrates effective classification application for early PDR prediction related to public infrastructure projects.

Taiwan has legally supported PPP projects for more than ten years. The National PPP Taskforce of the Taiwan Public Construction Commission (TPCC) is generally responsible for nationwide policies and in some cases provides advice about provisions for individual projects. Engineering departments and local governments are typically responsible for PPP project delivery. To achieve effective control of diverse projects under current workloads and to design proactive dispute management strategies, knowledge of possible PPP project dispute resolutions before disputes occur is essential to providing the governmental PPP Taskforce with information about future countermeasures. Additional preparation is generally beneficial once a dispute occurs by reducing the effort, time, and cost to multiple parties during dispute settlement. PPP projects involve devoted stakeholders, including a promoter (government), private investors, and financial institutions. Due to the high risks associated with the construction industry, repeated challenges for stakeholders can result in project delays, budget overruns, and poor construction quality during the implementation, construction, operating, and transfer phases. Although numerous studies (Abednego and Ogunlana, 2006, Cheung, 1999, Cheung et al., 2002, Gebken and Edward Gibson, 2006 and Jones, 2006) demonstrated that an efficient, effective, and fair dispute resolution process is essential for PPP project success, this study focuses on identifying warnings of potential dispute resolutions prior to project initiation. The proposed classification methods provide governmental authorities with the information needed to design proactive measures during project preparation and the phase when a dispute occurs. Many PPP projects initiated during the last decade have failed due to disputes occurring in the build, operate, and transfer (BOT) phases. According to the TPCC, the dispute rate was 23.6% during 2002–2009 (PCC, 2011). The most common processes for handling disputes are mediation/negotiation and non-mediation procedures. Non-mediation procedures include arbitration, litigation, and administrative appeals. In Taiwan, up to 84% of PPP projects are settled by mediation or negotiation within only 1–9 months after disputes occur (PCC, 2011). Notably, arbitration or litigation costs all parties considerably more time and money when a mediated agreement cannot be reached. This study applies multilabel classification models to early predict PPP dispute likelihood and potential resolutions, thereby alleviating the future adverse effects of disputes on project delivery, operation, and transfer from a governmental perspective. First, this study acquired historical dispute data for PPP projects started during 2002–2009 to establish functional relationships between project characteristics and their corresponding dispute resolutions. Differing from conventional construction project disputes, PPP project disputes may occur during the building phase, as well as during the operating, renting, or transfer phases. Thus, a second modeling phase with only dispute cases was implemented to identify the possibility of dispute resolutions under a set of known project attributes, dispute items, and the phase in which a dispute occurs. The rest of this paper is organized as follows. Section 2 thoroughly reviews artificial intelligence literature and its application in predicting construction claims and litigation outcomes. Section 3 then presents the research methodology and evaluation methods, respectively, providing a theoretical basis for classification models adopted in subsequent investigations. Section 4 describes the project dispute database and compares model performance based on classification techniques. Conclusions are finally drawn in Section 5, along with recommendations for future research.

This work has predicted potential dispute resolutions using multilabel classification and ensemble models. With known project attributes in different phases, the models can be developed to provide decisive information. For governmental agencies, the advantages of an early warning of dispute likelihood include reducing time and effort needed to prepare a rule set to prevent disputes by improving project understanding among project parties. This study used a number of different classifiers and the three best performing models and combined these models to predict possible dispute resolutions. Although the SVMs (91.58%), TAN (84.65%), and C5.0 (83.66%) in Phase I were the most accurate training models, the Exhaustive CHAID (83.82%), QUEST (82.99%), and C5.0 (82.57%) were the three best test models. The best test models were then utilized to develop ensemble models that achieved a classification accuracy of 84.65%. The best individual test models in Phase II were the SVMs (66.67%), CART (69.05), and Exhaustive CHAID (66.67) models, which were better than chance. However, the ensemble technique in Phase II did not improve classification accuracy, as it did in Phase I. Future work can investigate whether a fine tune of model parameter settings or a hierarchical ensemble approach combining multiple classifications, regression, and clustering techniques in a parallel or series improves model performance. Integrating proactive strategy deployment and preliminary countermeasures are also worthy of further study of early-warning systems for PPP project disputes. Based on dispute cases, such a system is needed to predict which dispute category and which resolution method are likely to be used at various phases of a project’s lifecycle by mapping hidden association rules.