Predicting Disputes in Public-Private Partnership Projects: Classification and Ensemble Models

Abstract

Proactively forecasting disputes in the initiation phase of public-private partnership (PPP) projects can considerably reduce the effort, time, and cost of managing potential claims. This comprehensive study compared classification models for PPP project dispute problems. Performance comparisons included four machine learners, four classification and regression trees, two multivariate statistical techniques, and combinations of techniques that have performed best according to a historical database. Experimental results indicate that an ensemble technique (i.e., SVMs+ANNs+C5.0) provides better cross-fold prediction accuracy (84.33%) compared with all other individual classification models. Notably, SVM (support vector machine) is the best single model for classifying dispute propensity in terms of overall performance measures. This study demonstrates the efficiency and effectiveness of data-mining techniques for early prediction of dispute propensity in PPP projects pertaining to public infrastructure services. The modeling results provide proactive-warning and decision-support information needed for managing potential disputes before disputes occur.