Early Dropout Prediction System for Secondary Schools in West Africa Using Machine Learning
West Africa has a secondary school dropout rate of 42%, the highest globally. In Ghana, Nigeria, and Senegal combined, over 3.2 million students abandon secondary education annually. School administrators currently …
Oluwasegun Odesola May 06, 2026 5 views 0 reactions
19/20
4Us Score
Problem Description
West Africa has a secondary school dropout rate of 42%, the highest globally. In Ghana, Nigeria, and Senegal combined, over 3.2 million students abandon secondary education annually. School administrators currently identify at-risk students only after they have already stopped attending — too late for intervention. No early warning system exists that works without internet connectivity, expensive software licences, or data science expertise. Teachers make dropout risk decisions based on intuition alone, missing 67% of at-risk students before it is too late.
Context
UNESCO data shows that each year of missed secondary education reduces lifetime earnings by 8-10% in Sub-Saharan Africa. Existing EdTech solutions (Tableau, PowerSchool) cost $15,000-80,000 annually per school district, making them inaccessible to public schools operating on $200 annual IT budgets. Government intervention programmes exist but target students already confirmed as dropouts rather than those at risk. The combination of late detection, unaffordable tools, and teacher overload creates a system where prevention is structurally impossible.
Target Audience
Secondary school headteachers, class teachers, and district education officers in West African public schools with limited IT infrastructure, no data science capability, and annual technology budgets
19/20
4Us Score
🤖 AI Validated
Validation
Education
Category
4Us Problem Worthiness Score
1️⃣ Unworkable
5/5
100%
Current dropout identification is entirely reactive. School registers are paper-based in 78% of West African public schools. Where digital records exist, no analytical tools process them. Teachers managing 60-80 students per class cannot manually track attendance trends, grade trajectories, and engagement signals simultaneously. The existing approach is structurally unworkable — it guarantees late detection by design.
2️⃣ Unavoidable
4/5
80%
Dropout pressure in West Africa is driven by poverty, early marriage, child labour demand during harvest seasons, and school fee structures. These root causes cannot be eliminated in the short term. However, early identification enables targeted interventions — fee waivers, counselling, flexible attendance — that retain students who would otherwise leave. The problem is unavoidable but early warning makes intervention possible.
3️⃣ Urgent
5/5
100%
West Africa's demographic bulge means 60% of the population is under 25. The window to educate this generation is closing. Students who drop out by age 15 rarely return. Ghana's Free Senior High School policy (2017) increased enrolment by 34% but dropout rates remain unchanged because no retention infrastructure exists. The urgency is compounded by a 2030 UN SDG4 deadline for universal quality education that West Africa is currently on track to miss by 40 years.
4️⃣ Underserved
5/5
100%
Every commercial early warning system requires cloud connectivity, SQL database expertise, or annual licensing fees that public West African schools cannot afford. Academic research exists on dropout prediction in US and European contexts but none validated on West African student data with its specific cultural, economic, and seasonal dropout patterns. Harvest season dropout spikes, school fee deadline correlations, and gender-specific risk factors in West Africa are entirely absent from existing models.
Total Score: 19/20
(95% on rubric scale)
— Decision:
✅ ACCEPT - Problem worth solving
Evidence Quality
0.0/10
⭐ Tier 1: 0📊 Tier 2: 0📄 Tier 3: 0💬 Tier 4: 0
Methodology
We built DropoutGuard, a random forest classifier trained on 6 years of anonymised student records from 47 public secondary schools across Ghana and Nigeria. The model uses 14 input features available in standard school registers: attendance rate per term, grade trend across 3 terms, number of fee payment delays, distance from school, gender, age relative to class, number of siblings enrolled, and 6 teacher-reported engagement scores. The system runs entirely as an offline Excel-compatible CSV tool, requiring no internet, no installation, and no technical training beyond basic spreadsheet use. Predictions are generated in under 3 seconds per student cohort on any laptop manufactured after 2010.
43,847 student records spanning 2018-2023 from 47 schools across Accra (Ghana) and Lagos (Nigeria). Data collected under MoU with Ghana Education Service and Lagos State Ministry of Education. Full ethics approval from University of Ghana IRB. Records anonymised at source — no names or national ID numbers included. Dropout confirmed via official school leaving records cross-referenced with district registers. Class imbalance (23% dropout, 77% retained) handled via SMOTE oversampling.
DropoutGuard achieved 88.3% accuracy, 84.7% precision, and 91.2% recall on the held-out test set of 8,769 student records. Compared to teacher prediction accuracy of 33% on the same cases, the model represents a 2.7x improvement. In a 6-month pilot across 8 schools in Accra, 847 at-risk students were identified in the first week of term. Of these, 612 received targeted interventions. End-of-term retention rate for flagged students who received intervention was 79%, compared to 31% historical retention for equivalent risk profiles. False positive rate of 15.3% was deemed acceptable by headteachers — identifying a non-at-risk student for a welfare check carries no negative consequences.
Key Findings
1. Attendance rate in term 1 was the single strongest predictor (feature importance 0.34), outperforming grades, fee payment, and all other variables combined.
2. Harvest season (October-November) created a distinct dropout signature detectable 6 weeks in advance through attendance micro-patterns not visible to teachers managing large classes.
3. Gender interacted significantly with distance — girls living more than 4km from school showed 3.1x higher dropout risk than boys at equivalent distance, revealing a safety-related barrier invisible in aggregate statistics.
4. The Excel interface was adopted without training by 94% of teachers in the pilot, validating the no-technical-expertise design goal.
5. Total deployment cost per school was $0 beyond existing hardware — the model runs on any laptop already present in the school office.
Limitations
1. Geographic scope: Training data covered only urban and peri-urban schools in Accra and Lagos. Rural school dropout patterns, which involve longer distances, seasonal agricultural labour, and different fee structures, were not represented and may reduce accuracy in those contexts.
2. Feature availability: The model requires 6 weeks of term data before generating reliable predictions. Week 1 and 2 predictions have accuracy of only 61%, limiting very early intervention in the first fortnight of term.
3. Teacher engagement scores: 3 of the 14 features depend on teachers completing a brief weekly engagement form. In the pilot, 31% of teachers completed this inconsistently, reducing model accuracy for those classrooms to 74%.
4. Language and literacy: The current interface is English-only. In Francophone West Africa (Senegal, Côte d'Ivoire) and in schools where teachers are more comfortable in local languages, adoption barriers exist that were not addressed in this version.
Product: DropoutGuard School Risk Dashboard
A zero-installation offline prediction tool for secondary school headteachers. Teachers enter or paste existing register data into a CSV template. DropoutGuard returns a ranked risk list with confidence scores and recommended intervention categories in under 3 seconds. No internet, no licence fee, no technical expertise required. Designed to run on any Windows or Mac laptop already present in a West African school office.
AI Validation:
✅ Completed May 06, 2026
Confidence:
94.0%
Reviewer Feedback:
AI Assessment: This is a well-executed research project that tackles a significant educational challenge in West Africa with a methodologically sound machine learning approach and demonstrates substantial improvement over current practices. The solution is thoughtfully designed for the target environment's constraints and shows strong empirical results with honest reporting of limitations.
Recommendation: ACCEPT
Strengths:
✅ Addresses a critical real-world problem with clear impact - 42% dropout rate and 3.2M affected students annually in West Africa
✅ Strong methodology with substantial dataset (43,847 records across 47 schools over 6 years) and rigorous evaluation metrics
✅ Practical solution design considering local constraints - works offline, uses standard school data, affordable for $200 IT budgets
Areas for Improvement:
💡 Expand geographic scope beyond urban areas to include rural schools where dropout patterns may differ significantly
💡 Provide more detailed information about the intervention strategies tested and their specific effectiveness rates
Passing Checks:
✓ Exceptional 4Us score: 19/20
✓ 4 dimensions score ≥3
✓ Detailed problem description
✓ Comprehensive methodology
✓ Detailed results description
✓ Honest limitations acknowledged
✓ Code repository provided
✓ Demo link provided
What this means
🤖 AI-verified - passed automated quality checks
⚡ Fast validation for basic quality standards
📊 May receive human review for higher tier promotion
No comments yet. Be the first to share your thoughts!
S
Oluwasegun Odesola
@segreen
I'm a researcher and data scientist at the intersection of artificial intelligence, algorithmic fairness, Financial and educational technology, with a …
Discussion (0)
No comments yet. Be the first to share your thoughts!