05 AI Big Data

Leveraging Artificial Intelligence and Big Data for IFAD 2.0

Accelerating knowledge generation for data-driven decision making

In order to unlock the potential of AI to accelerate knowledge generation and strengthen data driven decision making in IFAD, a multi-disciplinary team of economists, data scientists and social scientists used Machine Learning to extract insights from IFAD investments. This allowed systematic reviews to uncover the impact of key interventions, and the development of prediction models for performance and impact. As IFAD12 brings fewer, more focused, and larger investments, along with doubled impact and sustainability, a comprehensive picture of IFAD’s portfolio can support targeted interventions and strategic objectives.

“To support the effectiveness of IFAD interventions, we aimed to create a global picture using different data sources to map IFAD investments and understand where evidence gaps remain. Predictive analytics represents a new way to accelerate knowledge generation and inform policy decisions.”

Alessandra Garbero, Senior Economist, IFAD

RESULTS ACHIEVED

Project data from IFAD’s entire investment portfolio for over 30 years collected, organized and analysed
Text mining and modelling on more than 2,300 documents in four languages
Experimental software developed to analyse investments at project level, including a tool to standardise cost tables and a user-friendly search engine to display aggregated analysis and visualisations
Collaboration with Cornell University established to leverage an advanced machine learning model designed specifically for agriculture
Specific classes of interventions and outcomes identified; taxonomy created to analyse mainstreaming themes and Sustainable Development Goals
Topics detected through text mining combined with disbursement data, ratings and World Development Indicators for over 120 countries to create a prediction model for probability of project success and performance
Impact assessment data used to create a prediction model for probability of programme success and impact of household and project characteristics

POTENTIAL IMPACT

User-friendly dashboard envisioned as a final product that could integrate predictive analytics with data search and visualisation features to support project cycle through ex-ante predictions of performance and likelihood of positive effects
Results synthesis provided by systematic reviews and meta-analysis can complement project-level impact assessments where IFAD-specific evidence is lacking, as well as identify under-evaluated areas where IFAD should conduct more impact assessments
Support provided to IFAD’s ICT4D strategy through an integrated, machine-driven approach to analyse project documentation, predict impact and strengthen knowledge management by using new technologies to leverage existing data sources to answer new questions
Advancement of IFAD11 commitments and targets by accelerating project-level data analysis, enabling more regular reporting on the mainstreaming themes and SDGs against strategic outcomes

CHALLENGE AREAS

Short span of the project meant there was little time to collect and prepare the datasets, as well as to carry out pre- and post-tests, and to explore different models
Data collection was a major challenge, as there was not a central location from which to extract all documentation and documents were inconsistent
As IFAD reports are written in English, French, Spanish and Portuguese, we had to ensure all analyses considered multi-language models

Recommendations for next steps
(from the project team)

Refined

We recommend refining, and then scaling up. More time and further data collection are necessary to improve the categorization of interventions and the prediction algorithms. While the team achieved a strong conceptual framework for project performance prediction, we require more time to clean the data and improve the technicalities of the model, as well as to integrate the various sources and tools the project has created. We recommend adding more project-level documentation and then extending the analysis to the entire IFAD programme of work (including grants), which would not only deliver more consistent insights, but also help calibrate the prediction models.

Lessons learned
(from the project team)

A key lesson from this initiative is that effective machine learning is an iterative process that requires large amounts of data and resources to explore various approaches. It took extensive time to clean the data, given inconsistencies across documents. Project reports also contain a lot of extraneous information, which makes the data noisy and sometimes difficult to isolate what has really taken place. Nevertheless, this was an exciting project that allowed us to capitalise on existing data to uncover new patterns and gain additional knowledge. The applications for machine learning employed in this project can greatly support IFAD’s Development Effectiveness Framework, especially with knowledge generation, efficiency in corporate reporting, and building an evidence base to inform policy and the design of successful projects.

Back

The web page was made with Mobirise