Predicting policy funding allocation with Machine Learning

Allocating funds through competitive opportunities is a core tool of place-based development policies, as it can generate economic benefits and support the revitalisation of ‘left-behind’ territories. By relying on Machine Learning (ML) techniques, this paper investigates the predictability of actors expected to benefit from EU development funding over the 2014–2020 period in Italy. We implemented eight different ML classification algorithms and Random Forest, followed by Extreme Gradient Boosting, and Support Vector Machine emerged as the most predictive. The results show that it is possible to make out-of-sample predictions and diagnose the precise factors influencing fund allocation, such as territorial attributes, economic dimensions, and production specialisation. Knowing in advance potential winners of the calls can help design tailored territorial, and even sectorial, public policies to address the obstacles to local development and green transition, and to efficiently distribute resources within the policy framework. This evidence contributes to the reflection launched by the Commission on the future of the competitiveness of the EU. • This study uses ML to predict EU fund allocation in Italy (2014–2020) for policy design. • Random Forest, XGBoost, and SVM are the best models for predicting fund outcomes. • Funding success depends on territorial traits, economic factors, and applicant data. • The method aids policymakers and can be adapted to other EU funding schemes.