Hide/Show Apps

Development of dispute prediction and resolution method selection models for construction disputes

Ayhan, Mura
Construction industry is overwhelmed by increasing number and severity of disputes proving that current practices are insufficient in avoidance. This research argues that in order to forestall and mitigate construction disputes, prediction models should be developed by utilizing machine learning algorithms. The research suggests developing three distinct models; (1) dispute occurrence prediction model, (2) potential compensation prediction model, and (3) resolution method selection model. For this reason, an extensive literature review is conducted to identify input variables that impact dispute occurrence, compensation, and resolution method selection. Findings of the literature review is used to develop a conceptual model that involves attributes related to project, parties, dispute, and resolution method characteristics along with attributes related to changes, delays, and knowledge on resolution methods. Then, a questionnaire is designed based on the conceptual model to collect empirical data from decision-making authorities. Chi-Square tests of association is performed on collected datasets to reveal the significance of associations between inputs and outputs. Insignificant attributes are eliminated and finalized prediction models are developed. These models are tested by using alternative single and ensemble machine learning algorithms to obtain the best classifiers. 10-fold cross-validation results with ten repeats showed that dispute occurrence prediction model achieved 91.11% average prediction accuracy while potential compensation prediction model achieved 80.61% average accuracy and resolution method selection model has 89.44% average classification accuracy. These promising results show that proposed models can be beneficial for management personnel by supporting the decision-making process in future disputes based on data from past disputes.