1 of 12

AI in Education in Emergencies: Transforming Monitoring, Evaluation, and Learning (MEL) for Refugee Pathways��A Case Study of the AHEEN RDP Pathways Program

Presenter: Immaculate Wanjiru Kimani

Date: 22-25 July 2025

Affiliation: African Higher Education in Emergencies Network (AHEEN), University of Nairobi

2 of 12

Higher Education Refugee Learners

3 of 12

The Global Refugee Education Crisis

Sub-Saharan Africa hosts nearly 30% of the world's displaced population.
Only 6% of refugee youth access higher education globally, compared to 40% worldwide.

Barriers:

Recognition of prior qualifications
Financial hardship
GBV and cultural barriers -> Girls

4 of 12

Introducing the AHEEN RDP Pathways Program

5 of 12

Challenges with Traditional MEL Systems in Refugee Education

Rely on manual processes that cannot capture the complexity of multidimensional interventions.
Lack mechanisms to track the 78% of refugee students who disappear between secondary completion and tertiary enrolment, missing intervention opportunities.
Without AI-enhanced analytics, programs cannot proactively identify at-risk students until academic failures occur, particularly for targeted groups like women and persons with disabilities.

6 of 12

The AI-Driven MEL Framework: Meeting Key Needs

This study develops an AI-driven MEL framework capable of:

Real-time integration of cross-pillar data (athletics → education → labour).
Addressing transition blindspots and allows proactive identification of at-risk students.
Adaptive learning for continuous program improvement: predictive > reactive.
Offers a scalable solution to the "last mile problem" in refugee education, transforming access into meaningful outcomes.

7 of 12

Key Usages of AI in MEL and Education��

Integrates fragmented data from athletics participation, academic performance, and tertiary transition points.
Machine learning techniques analyse quantitative indicators and qualitative narratives to identify success predictors and at-risk students.
Natural Language Processing (NLP) for sentiment and theme extraction from unstructured feedback (e.g., stakeholder interviews).
AI tools can automate admissions, enrolment, and financial aid processing.
LangChain-integrated GPT-4 for automated report drafting and GenAI for PDF format reports.

8 of 12

AI-Enhanced MEL: Core Methodologies and Tools��

Objective	Technique	Tools
Trend Prediction	Prophet time-series modeling	Python, SQL
Dropout risk analysis	Survival analysis (Kaplan-Meier)	Scikit-learn
Qualitative insight mining	NLP (Sentiment/theme extraction)	GPT-4.0 mini /Llama-3
Bias detection	Statistical parity testing	SciPy, Hugging

9 of 12

AI-Enhanced MEL: System Workflow��

10 of 12

Ethical Considerations: Ensuring Fair and Trustworthy AI

Data Privacy & Security: Anonymisation and de-identification of student identities. Role-based data access controls in the DBMS.

Trustworthy Algorithms: Ensuring security, reliability, validity, transparency, and accuracy of AI algorithms and interpretability of outputs.

Fairness and Equity: Addressing algorithmic bias present in historical data that can lead to unfair outcomes for marginalised groups (e.g., based on gender, disability status)

Mitigation: Regular fairness audits using SHAP (SHapley Additive exPlanations) values. Techniques like re-weighting training data if bias is detected.

Human Oversight & Accountability: AI systems should augment, not replace, human judgment. Clear accountability structures are needed for AI-driven decisions.

Stakeholder Involvement & AI Literacy: Educators, students, and administrators must be adequately involved in AI system design to align with real-world educational needs. AI literacy for all involved is a priority.