Developing a UAE-Based Disputes Prediction Model using Machine Learning

Abstract

Disputes are a major phenomenon in the construction industry around the world that stems from unsettled disagreements between project stakeholders. These types of disagreements can take place in any project regardless of its size and properties. The UAE is no different. To resolve these disputes, a substantial amount of money and time must be allocated which might cause the project to collapse. As a result, proactive construction management is needed to prevent this issue from arising in the first place. The aim of this research is to develop prediction models for construction disputes, thereby providing early insights to the stakeholders, and thus making precautionary measures that can prevent these disputes from taking place during the project execution. Initially, a literature review was performed to gather the input parameters needed for the prediction models. These parameters were verified by industry experts using a preliminary survey to find out the most important ones. The top three parameters were found to be the general experience and competence of the contractor, the project size, and the level of contract readiness. Moreover, another survey was administered in order to acquire actual project data that will be the input for the proposed prediction models. The sample size was 79 projects, where 67% of these projects faced disputes. These models will be able to predict dispute occurrence, the number of disputes, the impact of disputes on time and cost, the dispute resolution procedure, as well as the time and cost of the disputes resolution. Additionally, three different Machine Learning (ML) algorithms, Artificial Neural Networks (ANN), Support Vectors Machine (SVM), and Random Forests, were used to run these models and perform predictions. It was found that SVM and Random Forests provided better results in terms of accuracy in all of the seven models. Most of the models were well-performing since the testing accuracies lies within the 70-90% range, with disputes resolution duration prediction model even exceeded the 90% mark. Furthermore, as demonstrated in the case study, these models were successful in predicting the different aspects of disputes and demonstrated that they can be implemented in future projects to achieve disputes mitigation.