Design of a Data-driven Intervention Dashboard for SDG Localization

Pooja Bassin, Abraham G. K., Srinath SrinivasaIJCAI 2024

Abstract

Achieving the United Nations' Sustainable Development Goals (SDGs) requires localizing global targets to specific regional contexts. This paper presents a data-driven dashboard developed for Karnataka, India, that moves beyond descriptive reporting to predictive and prescriptive modeling. By identifying key drivers of maternal mortality and incorporating a novel stability metric, the tool provides actionable policy recommendations and budget allocation strategies to reach health targets while ensuring regional equilibrium.

Introduction

The success of the Sustainable Development Goals (SDGs) hinges on "localization"—the adaptation of global strategies to fit unique socio-economic and cultural contexts. In the Indian state of Karnataka, district-level conditions vary significantly. While Open Government Data (OGD) is plentiful, policymakers often lack tools to translate this information into actionable interventions. Existing dashboards typically function as reporting tools, showing historical trends without offering guidance on how to achieve specific targets, such as reducing maternal deaths (MD).

Methodology

The researchers developed a three-tiered modeling approach integrated into a decision-support dashboard.

Predictive and Prescriptive Modeling

Using multivariate linear regression, the team identified four primary drivers of maternal mortality from an initial set of 115 attributes: Caesarean sections, institutional births, female leprosy cases, and female TB deaths. The prescriptive phase allows users to set a target for maternal deaths (e.g., reducing the state average from 19.42 to 10). The model then calculates the necessary changes in the four key drivers based on real-world feasibility.

Prescriptive modeling interface showing target maternal deaths and required changes in key drivers.

Stability and Stress Modeling

A core innovation of this work is the "Stability Score." Interventions that improve one district while widening the gap with its neighbors can create regional instability. The dashboard maps districts into quadrants based on "Impact" and "Stability," helping identify where improvements are most sustainable.

Impact vs. Stability plot identifying high-impact and low-stability districts.

Budget Allocation

The tool bridges data science and public finance by linking statistical sensitivity parameters to budget formulas. It suggests how funds should be distributed across government schemes—such as health facility improvements or tuberculosis programs—to maximize the impact on maternal health.

Datasets

The dashboard integrates data from five major OGD sources: Karnataka At a Glance (KAG), National Family Health Survey (NFHS-4), Karnataka Health Profile, Men and Women in Karnataka, and the eJanMa digital registration system.

Key Results and Insights

The model demonstrated that focusing on institutional deliveries and reducing female TB deaths could lower the state average for maternal deaths from 19.42 to 10. The stability analysis highlighted districts like Dharwad (DWD) as high-impact areas that require careful management to ensure long-term sustainability. Furthermore, the budget model prioritized funding for Health Facilities (HF) as the most effective lever for change.

Conclusion

While currently limited by the granularity of public district-level data, this dashboard represents a significant shift from descriptive monitoring to prescriptive policy design. Future work will expand the tool to include cross-domain indicators and explore lateral impacts between different SDGs, such as the link between education and health outcomes.

Summary updated

BibTeX
@inproceedings{bassin2024design,
  author = {Pooja Bassin and Abraham G. K. and Srinath Srinivasa},
  title = {{Design of a Data-driven Intervention Dashboard for SDG Localization}},
  booktitle = {International Joint Conference on Artificial Intelligence},
  year = {2024},
}
Figure