MWI-002

MIRA – Machine Learning Food Security Forecasting for Anticipatory Action (Southern Malawi)

Download PDF
Malawi Sub-Saharan Africa Low income Pilot / Controlled Trial Phase Confirmed

Catholic Relief Services (CRS) – lead implementer

At a Glance

What it does Prediction (including forecasting) — Trend and shock forecasting
Who runs it Catholic Relief Services (CRS) – lead implementer
Programme United in Building and Advancing Life Expectations (UBALE) – MIRA sentinel site monitoring component
Confidence Confirmed
Deployment Status Pilot / Controlled Trial Phase
Key Risks Data-related risks
Key Outcomes Best-performing model (Random Forest) achieved F1 score of approximately 81% and accuracy of approximately 83% in predicting household food security outcomes using rCSI threshold-16 dichotomisation with 20 AI-selected predictor variables plus historical food security scores.
Source Quality 4 sources — News article / media, Working paper / technical note, Report (multilateral / development partner), +1 more

The Measurement Indicators for Resilience Analysis (MIRA) system is a machine-learning-enabled food security forecasting tool developed by Catholic Relief Services (CRS) in collaboration with Cornell University's Charles H. Dyson School of Applied Economics and Management, with subsequent research collaboration with Microsoft's AI for Good Lab. The system was designed to forecast near-term household food security outcomes up to approximately four months ahead, enabling anticipatory action, targeting of assistance, and early warning within humanitarian and social assistance programming in southern Malawi (Gholami et al., 2022, Data & Policy, Vol. 4).

MIRA was conceived and piloted within the context of the United in Building and Advancing Life Expectations (UBALE) programme, a five-year, USD 63 million USAID Food for Peace-funded Development Food Assistance Program managed by CRS-Malawi. UBALE served approximately 250,000 vulnerable households across 284 communities in three of Malawi's poorest and most disaster-prone districts: Chikwawa, Nsanje, and Rural Blantyre (CIO, 2019). The MIRA data collection protocol was officially launched in May 2016, beginning with a baseline evaluation, and was subsequently piloted over three months in Chikwawa District before expansion (CIO, 2019).

The MIRA protocol employs 'embedded enumerators' — community members residing in selected sentinel sites — who administer a high-frequency household survey on a monthly basis to the same panel of households. This longitudinal data collection approach captures information on demographics, livelihood and economic status, shock exposure history (including new shocks each month), food security outcomes measured through the reduced Coping Strategies Index (rCSI), household well-being indicators, and household capacities such as land assets, livestock ownership, education levels, and disability status (CIO, 2019). Surveyed households were originally selected using flood exposure maps from the Dartmouth Flood Observatory, and results were organised by flood risk groups and resilience variables tracking vulnerability to natural disasters (CIO, 2019).

The machine learning component of MIRA was developed through a research collaboration between CRS and Microsoft's AI for Good Lab. The peer-reviewed study published in Data & Policy (Gholami et al., 2022) benchmarked multiple supervised machine learning algorithms on the MIRA sentinel site data, including Random Forest, neural networks, and classical statistical models. The study used a binary classification approach to predict household food security outcomes, with the rCSI score dichotomised at a threshold of 16 to distinguish food-secure from food-insecure households. The Random Forest model outperformed other algorithms when focusing on predictors of community-level vulnerability. The best-performing model configuration achieved an F1 score of approximately 81 percent and an accuracy of approximately 83 percent in predicting food security outcomes, using historical food security scores combined with 20 predictor variables selected through an AI explainability framework (Gholami et al., 2022).

To ensure model interpretability, the researchers applied Shapley Additive Explanations (SHAP), a game-theoretic framework that quantifies the contribution of each predictor feature to the final prediction. SHAP analysis revealed that geographic location and self-reported welfare indicators were the most important predictors of food insecurity at the community level (Gholami et al., 2022). This cross-validation and model benchmarking approach, combined with SHAP-based feature importance analysis, provided a degree of interpretability and transparency to the forecasting outputs.

Operationally, early warning information generated through MIRA has been shared with local village development committees to guide community-level responses to emerging food security threats (CIO, 2019). MIRA data delivered to communities has enabled them to inform local government actors of problem hotspots, such as early reports of Fall Armyworm infestations, and has helped small-scale aid organisations, including church groups, to design localised interventions (Oxford Policy Management, 2022). During the COVID-19 pandemic, MIRA data was also used in research that revealed the disproportionate impact of the pandemic on already food-insecure households (Oxford Policy Management, 2022).

Microsoft awarded CRS an AI for Humanitarian Action Grant to develop a plug-and-play version of the MIRA solution, with the goal of enabling any CRS country programme to identify food insecurity drivers and pre-position aid more efficiently (CIO, 2019). The MIRA approach has been implemented or piloted in additional countries, including Madagascar and Ethiopia. CRS won a FutureEdge 50 Award for the MIRA system's emerging technology applications (CIO, 2019).

The system operates in a human-on-the-loop (HOTL) paradigm: model outputs inform decisions by programme staff and community actors, but there is no evidence of automatic rights-affecting actions being triggered by algorithmic outputs. The decision criticality is assessed as low, as the system provides informational and triage-level advisory analytics rather than determining individual benefit eligibility or entitlements. However, no public documentation of a formal Data Protection Impact Assessment, standard operating procedures, or governance artefacts specific to the MIRA Malawi implementation has been located.

Classifications follow the DCI AI Hub Taxonomy. Hover over field labels for definitions.

Social Protection Functions

Implementation/delivery chain
Assessment of needs/conditions + enrolment primaryMonitoring and evaluation Outreach/communications/sensitisation
SP Pillar (Primary) The social protection branch: social assistance, social insurance, or labour market programmes. Social assistance
Programme Name United in Building and Advancing Life Expectations (UBALE) – MIRA sentinel site monitoring component
Programme Type The type of social protection programme, classified under social assistance, social insurance, or labour market programmes. View in glossary Emergency Cash Transfers
System Level Where in the social protection system the AI is applied: policy level, programme design, or implementation/delivery chain. View in glossary Implementation/delivery chain
Programme Description UBALE was a five-year, USD 63 million USAID Food for Peace-funded Development Food Assistance Program managed by CRS-Malawi, serving approximately 250,000 vulnerable households across 284 communities in Chikwawa, Nsanje, and Rural Blantyre districts. MIRA was the high-frequency resilience monitoring and food security forecasting component embedded within UBALE.
Implementation Type How the AI output is produced: Classical ML, Deep learning, Foundation model, or Hybrid. Affects validation, compute requirements, and governance profile. View in glossary Classical ML
Lifecycle Stage Current stage in the AI lifecycle, from problem identification through to monitoring, maintenance and decommissioning. View in glossary Integration and Deployment
Model Provenance Origin of the AI model: developed in-house, adapted from open-source, commercial/proprietary, or accessed via third-party API. View in glossary Adapted from open-source
Compute Environment Where the AI system runs: on-premise, government cloud, commercial cloud, or edge/device. View in glossary Not documented
Sovereignty Quadrant Classification of data and compute sovereignty: I (Sovereign), II (Federated/Hybrid), III (Cloud with safeguards), or IV (Shared Innovation Zone). View in glossary Not assessed
Data Residency Where the data used by the AI system is stored: domestic, regional, or international. View in glossary Not documented
Cross-Border Transfer Whether data crosses national borders, and if so, whether documented safeguards are in place. View in glossary Not documented
Decision Criticality The rights impact of the decision the AI supports. High criticality requires HITL oversight; moderate requires HOTL; low may operate HOOTL. View in glossary Low
Human Oversight Type Level of human involvement: Human-in-the-Loop (active review), Human-on-the-Loop (monitoring), or Human-out-of-the-Loop (periodic audit). View in glossary HOTL
Development Process Whether the AI system was developed fully in-house, through a mix of in-house and third-party, or fully by an external provider. View in glossary Mix of in-house and third-party
Highest Risk Category The most significant structural risk source identified: data, model, operational, governance, or market/sovereignty risks. View in glossary Data-related risks
Risk Assessment Status Whether a formal risk assessment, informal assessment, or independent audit has been conducted for this system. Not assessed

Impact Dimensions

Autonomy, human dignity and due process
Systemic and societal
  • Bias audit
  • Human oversight protocol
CategorySensitivityCross-System LinkageAvailabilityKey Constraints
Geospatial and remote sensing dataNon-personalLinks data across multiple systemsCurrently available and usedDartmouth Flood Observatory flood exposure maps used for initial household selection stratification by flood risk group; geographic location identified as top predictor via SHAP analysis
Survey and census dataPersonalSingle source (no linkage)Currently available and usedHigh-frequency monthly household surveys from MIRA sentinel sites administered by embedded community enumerators; captures demographics, livelihood, shock exposure, rCSI food security scores, household capacities (land, livestock, education, disability); panel design tracking same households monthly

Florance, M. (2019) 'Catholic Relief Services Leverages Machine Learning to Fight Hunger', CIO, 7 March. Available at: https://www.cio.com/article/196076/catholic-relief-services-leverages-machine-learning-to-fight-hunger.html (Accessed: 24 March 2026).

View source News article / media

Microsoft Research (2022) 'Food Security Analysis and Forecasting: A Machine Learning Case Study in Southern Malawi', Microsoft Research Publications. Available at: https://www.microsoft.com/en-us/research/publication/food-security-analysis-and-forecasting-a-machine-learning-case-study-in-southern-malawi/ (Accessed: 24 March 2026).

View source Working paper / technical note

Oxford Policy Management (2022) 'Can Better Data Help Policy-Makers and Communities Build Resilience?'. Oxford: Oxford Policy Management. Available at: https://www.opml.co.uk/insights/can-better-data-help-policy-makers-and-communities-build-resilience (Accessed: 24 March 2026).

View source Report (multilateral / development partner)

Gholami, S., Knippenberg, E., Campbell, J., Andriantsimba, D., Kamle, A., Parthasarathy, P., Sankar, R., Birge, C. and Lavista Ferres, J.M. (2022) 'Food Security Analysis and Forecasting: A Machine Learning Case Study in Southern Malawi', Data & Policy, 4, e38. doi:10.1017/dap.2022.24.

View source Academic journal article
Deployment Status How far the system has progressed into real-world operational use, from concept/exploration through to scaled and institutionalised. View in glossary Pilot / Controlled Trial Phase
Year Initiated The year the AI system was first initiated or development began. 2016
Scale / Coverage The scale and geographic or population coverage of the deployment. Sentinel sites across 284 communities in three districts (Chikwawa, Nsanje, Rural Blantyre) within the UBALE programme area serving approximately 250,000 households; MIRA sentinel panel is a subset of these communities.
Funding Source The source(s) of funding for the AI system development and deployment. USAID Food for Peace (UBALE programme); Microsoft AI for Humanitarian Action Grant (ML development)
Technical Partners External technology vendors, academic partners, or development partners involved. Microsoft AI for Good Lab (ML model development and research collaboration); Cornell University (MIRA protocol co-design); CRS internal MEAL unit (data collection and operational integration). ML models used standard off-the-shelf algorithms (Random Forest, neural networks) — no proprietary platform documented.
Outcomes / Results Best-performing model (Random Forest) achieved F1 score of approximately 81% and accuracy of approximately 83% in predicting household food security outcomes using rCSI threshold-16 dichotomisation with 20 AI-selected predictor variables plus historical food security scores. SHAP analysis identified location and self-reported welfare as top predictors. Demonstrates feasibility of short-horizon forecasting on sentinel data. Early warning information shared with village development committees. MIRA data enabled identification of Fall Armyworm hotspots and informed localised interventions. COVID-19 impact research conducted using MIRA data.
Challenges Sentinel-site sampling limits generalisability beyond monitored communities. Forecasting horizon limited to approximately four months. No documentation of routine programme activation or operational triggers based on ML outputs in Malawi. No public DPIA or governance artefacts specific to the MIRA Malawi implementation located. Hosting location and compute infrastructure not publicly documented.

How to Cite

DCI AI Hub (2026). 'MIRA – Machine Learning Food Security Forecasting for Anticipatory Action (Southern Malawi)', AI Hub AI Tracker, case MWI-002. Digital Convergence Initiative. Available at: https://socialprotectionai.org/use-case/MWI-002 [Accessed: 1 April 2026].

Change History

Created 30 Mar 2026, 08:40
by v2-import (import)