Classifying Compensations in Construction Disputes using Machine Learning Techniques

Abstract

It is highly probable to encounter disputes in construction projects and construction disputes are detrimental as they may lead to cost overruns and delays. Knowing the potential compensation with some certainty can avoid parties from extending their claims. Parties may benefit from decision support systems that help them to understand whether they can acquire any compensation and in what aspect depending on the disputed case characteristics. Within this context, it is hypothesized that compensations in construction disputes can be forecasted by using machine learning (ML) techniques. For this reason, findings of an extensive literature review were used to develop a conceptual model that identifies factors affecting compensations in construction disputes. Using the input variables from the conceptual model, a questionnaire was designed to collect empirical data from experts. Attribute elimination was performed on the collected data via Chi-square tests to reveal the associations between input variables and compensations. Insignificant attributes were eliminated to develop a classification model and the obtained model was experimented via alternative single and ensemble ML techniques. The best classification performance was obtained from the Naïve Bayes (NB) algorithm with an average classification accuracy of 80.61% when One-vs-All (OvA) decomposition technique was utilized. The conceptual model can guide construction professionals during dispute management decision-making and the promising results indicate that the classification model has the potential to identify compensations. This study can be used to mitigate construction disputes by preventing parties from resorting to unpleasant and inconclusive resolution processes.