Predictive modelling for contract selection in construction through data mining and machine learning: Insights from agency theory

The client-agent relationship is crucial for successful construction projects, but agency problems like adverse selection and moral hazard can weaken it if the appropriate contract type is not chosen. Previous efforts developed a multinomial logistic regression model to predict contract types based on agency theory indicators in construction projects. However, the model faced challenges with performance, complexity, data imbalance, and over-optimistic results. To address this gap, this research introduces a novel contract selection model grounded in agency theory. Using data mining and three machine learning algorithms–Random Forest, Multilayer Perceptron, and pruned Decision Tree–the study improves upon prior models. The Random Forest and pruned Decision Tree outperformed the previous model with accuracies of 74.5% and 72.3%, respectively, while the Multilayer Perceptron achieved just below 70%. However, all models are more reliable due to cross-validation. The pruned Decision Tree highlights five key variables: the final product's structure and function, the contractor's experience, commitment, neighboring conditions, and human resources. The models help principals and decision-makers select optimal contract types, mitigating agency problems and promoting collaboration in construction projects. This research advances Construction 4.0's digitalization goals through a data-driven approach and provides an interpretable Decision Tree for stakeholders without technical expertise.