Analyzing the differential impact of variables on the success of solicited and unsolicited private participation in infrastructure projects using machine learning techniques

Purpose – The study aims to examine the impact of key variables on the success of solicited and unsolicited private participation in infrastructure (PPI) projects using machine learning techniques. Design/methodology/approach – The data has information on 8, 674 PPI projects primarily derived from the World Bank database. In the study, a machine learning framework has been used to highlight the variables important for solicited and unsolicited projects. The framework addresses the data-related challenges using imputation, oversampling and standardization techniques. Further, it uses Random forest, Artificial neural network and Logistics regression for classification and a group of diverse metrics for assessing the performances of these classifiers. Findings – The results show that around half of the variables similarly impact both solicited and unsolicited projects. However, some other important variables, particularly, institutional factors, have different levels of impact on both projects, which have been previously ignored. This may explain the reason for higher failure rates of unsolicited projects. Practical implications – This study provides specific inputs to investors, policymakers and practitioners related to the impacts of several variables on solicited and unsolicited projects separately, which will help them in project planning and implementation. Originality/value – The study highlights the differential impact of variables for solicited and unsolicited projects, challenging the previously assumed uniformity of impact of the given set of variables including institutional factors.