Methodology

Development projects are inherently multidimensional.

A single activity may simultaneously address climate adaptation, improve agricultural productivity, strengthen governance institutions and invest in capacity building. Describing such interventions using only a traditional sector/subsector classification inevitably hides part of their complexity.

For this reason, the platform adopts a multidimensional analytical model rather than a single-axis taxonomy.

Instead of asking “What sector does this project belong to?”, the platform asks a broader set of questions:

Which environmental challenge does the project primarily address? What specific environmental objective is pursued? In which Intervention Area is the intervention implemented? Which transversal dimensions characterize the project?

This approach provides a richer and more realistic representation of development portfolios while preserving comparability across donors, countries and implementation sectors.

Every activity follows the same analytical workflow before becoming part of the dataset.

Official IATI Activity
│
▼
Data Normalization
│
▼
Structured Metadata Analysis
│
▼
DAC Sector Analysis
│
▼
Policy Marker Analysis
│
▼
Natural Language Processing
│
▼
Knowledge Base Matching
│
▼
Environmental Theme
│
▼
Thematic Focus
│
▼
Intervention Area
│
▼
Cross-cutting Flags
│
▼
Quality Assurance
│
▼
Analytical Dataset

Each step enriches the information available for analytical purposes while maintaining traceability to the original IATI activity.

Sources of Classification Evidence

The platform never relies on a single information source.

Instead, it combines multiple complementary sources of evidence to improve robustness and reduce the impact of incomplete or inconsistent reporting.

These sources include:

DAC purpose codes; OECD policy markers; structured IATI metadata; project titles; project descriptions; participating organisations; geographic information; normalized keywords; Natural Language Processing (NLP); rule-based analytical knowledge.

Combining these sources allows the platform to identify environmental relevance even when individual fields are incomplete or ambiguous.

The platform classifies every activity using four complementary analytical dimensions.

Unlike a traditional hierarchical taxonomy, these dimensions describe different aspects of the same intervention and therefore coexist rather than replace one another.

Environmental Theme

The Environmental Theme identifies the primary environmental or climate-related challenge addressed by the project.

This is the platform’s principal analytical dimension and provides the highest level of classification confidence.

Each activity is assigned to one primary Environmental Theme.

Within each Environmental Theme, projects are further classified into a more specific analytical category whenever sufficient evidence is available.

Thematic Focus supports detailed portfolio analysis and drill-down exploration.

When evidence is insufficient to identify a specific category confidently, the activity is assigned to Other.

This should be interpreted as “insufficient evidence for further specialization”, not as a classification error.

The Intervention Area identifies the development sector in which project activities are primarily implemented.

For example, a Climate project may operate mainly within Agriculture, Health, Water, Governance or Energy.

Separating environmental objectives from implementation domains provides a more realistic understanding of development interventions.

Many projects include transversal dimensions that span multiple Environmental Themes and Intervention Areas.

These dimensions are represented independently through Cross-cutting Flags.

Examples include:

• Governance
• Capacity Building
• Research
• Data & Information Systems
• Finance & Investment

A project may contain none, one or several Cross-cutting Flags.

A key principle of the platform is explainability.

Every classification is supported by identifiable analytical evidence rather than opaque machine-learning predictions.

Whenever available, Project Intelligence exposes the evidence contributing to a project’s classification, including structured metadata, environmental signals and supporting information extracted during processing.

This enables analysts to understand not only what the platform classified, but also why it reached that conclusion.

Different analytical dimensions are associated with different levels of confidence.

The Environmental Theme generally represents the most robust classification because it integrates evidence from multiple independent sources.

More detailed dimensions, such as Thematic Focus, require stronger textual evidence and are therefore intentionally conservative.

The presence of an Other category should not be interpreted as poor data quality. It simply indicates that the available evidence does not support a more specific classification with sufficient confidence.

Classification quality is continuously monitored through a dedicated quality assurance framework.

This includes:

automated regression testing; consistency validation; classification audits; false-positive analysis; false-negative analysis; iterative refinement of analytical rules.

This continuous validation process allows the analytical framework to evolve while maintaining consistency across dataset releases.

The analytical framework is designed to evolve over time.

As additional projects are analysed and new environmental patterns emerge, classification rules, thematic categories and supporting knowledge can be refined.

When methodological improvements are introduced, affected activities may be reprocessed so that the analytical dataset consistently reflects the latest validated methodology.

Design Principle

Classify Once, Analyse Everywhere

Every activity is classified only once using a consistent analytical methodology. The resulting multidimensional profile becomes the common semantic foundation for all components of the platform—including Explore, Maps, Multidimensional Analysis and Project Intelligence. This ensures that users obtain coherent and comparable results regardless of how they explore the dataset.