Empirical inference system for highway project delivery selection using fuzzy pattern recognition

Selection of a project delivery method in highway construction is a challenging task because of the many decision criteria involved. In addition to quantitative project attributes, the project delivery decision-making process also typically relies on qualitative measures such as subjective judgments and experts' opinions based on their experience with similar completed projects. Although current probabilistic methods provide a robust means to analyze quantitative data, they are not ideally suited for treating uncertainties encountered in qualitative data. To overcome the identified gap, this study investigated fuzzy pattern recognition, a mathematical technique based on fuzzy sets and fuzzy logic, to model a combination of quantitative and qualitative variables in highway project delivery selection. A fuzzy rule-based inference system was developed, trained, and tested using 254 empirical highway projects with particular project attributes including project type, project complexity, delivery risk, and cost performance. The proposed system was verified by performing a case project with the result of accurately recognizing the true project delivery method used and associated cost growth performance expectations. The flexibility of fuzzy membership functions in the proposed system helps leverage the evaluation of a combination of quantitative and qualitative variables in project delivery method selection in complex highway construction projects. In addition, this data-driven fuzzy inference system also allows for multiple decision scenarios based on the decision maker's judgments of delivery risks and project complexity. This study contributes to the body of knowledge by developing an empirical inference system to recognize possible patterns of delivery methods associated with cost growths for new highway projects. This study may assist highway agencies in making project delivery decisions based on project attributes, historical data, and their relevant experience.