ABSTRACT
Philanthropic support for international development makes up an important and growing element of development aid. The OECD collects and publishes data on private philanthropy for development. It classifies these activities using a standardized framework for classifying development projects. Until now, the task of classification has been accomplished through manual review of project descriptions, an onerous process that limits the amount of information that can be analyzed and introduces multiple reporting and classification biases. To expand the scope and improve the consistency of OECD data on philanthropy, this chapter introduces PHIL4DEV, a supervised machine learning model trained on OECD philanthropic data available as of 2019. PHIL4DEV uses a bag-of-words approach and XGBoost optimization to automate the classification of purpose codes for philanthropic activities. While the initial results of this model are promising, with a weighted micro-F1 score of 0.74, many improvements are needed to achieve a fully functional tool. The chapter concludes by discussing the limitations of the first version of the model, comparing its predictive power with more sophisticated algorithms, and the computational limitations it would face in handling large flows of users and data. These challenges will be addressed in a new version of PHIL4DEV to be developed in 2024.
