Explainable Predictive Factors for Inter-Agency Partnership Success

Development partnerships that overcome the intrinsic limitations of individual organizations are essential to achieve the goals of the 2030 Agenda for Sustainable Development. However, one in five strategic partnerships fails, incurring higher costs for lesser impact. Assuming that partnership performance is proportionate to co-financed project ratings, we used the corresponding evaluation reports and project ratings from multiple multilateral organizations to automate the identification and analysis of critical predictive factors for partnership success. This approach ultimately helps form more selective partnerships. Our automation engine integrates text mining, sentiment and semantic analysis, and supervised Machine Learning with explainable and opaque algorithms. To achieve this goal, we compared four alternative Machine Learning models. By applying Shapley's additive explanation values to trained models, we ranked factors by criticality, identifying their Top-10 rank. With bootstrap sampling, instead, we computed confidence intervals. Our rating predictions were rather good. Explainable AI models achieved a Matthews Correlation Coefficient of 0.749. Opaque algorithms reached 0.780. Both compare well with the current success rates of 81% for project co-financing partnerships and 75% for knowledge partnerships. Our prototype engine may help organizations improve partnership appraisal efficiency and improve collaboration. Future work will include refining factor formulations, deploying models with incremental learning, and introducing causality analysis.