🤖 AI Powered Data Analytics Lab
Apply AI concepts through hands-on exercises in predictive analytics, tool selection, and real-world scenarios
🎯 Predictive Analytics Practice
Learn to identify when and how to use predictive analytics in M&E contexts.
📚 Understanding Prediction Types
📈 Regression
Predicts: Continuous numerical values
Example: Predict income levels, attendance rates, crop yields
When to use: "How much?" or "How many?"
🏷️ Classification
Predicts: Categories or labels
Example: High/medium/low risk, dropout/non-dropout
When to use: "Which category?" or "Yes/No?"
📊 Time Series
Predicts: Future values based on historical patterns
Example: Disease outbreaks, seasonal trends, program uptake
When to use: Data collected over time
Interactive Exercise: Student Dropout Prediction
Context: An education program has 5 years of data on 10,000 students including attendance, test scores, family income, and dropout status. You want to predict which current students are at risk of dropping out.
Your Task: Analyze the dataset characteristics and determine the best predictive approach.
📊 Sample Dataset Preview
| Student ID | Attendance Rate | Test Score Avg | Family Income | Distance to School (km) | Dropout Status |
|---|---|---|---|---|---|
| 001 | 92% | 78 | $300/month | 2.5 | No |
| 002 | 65% | 52 | $180/month | 8.0 | Yes |
| 003 | 88% | 85 | $450/month | 1.2 | No |
| 004 | 58% | 48 | $150/month | 12.0 | Yes |
🎯 Key Questions to Answer
1. What type of machine learning is most appropriate?
2. Which features (variables) are likely most predictive of dropout?
Select all that apply by checking the boxes:
💡 Predictive Analytics Best Practices
- Start with clear objectives: What exactly are you trying to predict?
- Check data quality: Predictions are only as good as your data
- Validate models: Test on data the model hasn't seen before
- Monitor performance: Accuracy can drift over time as contexts change
- Combine with human judgment: AI predicts, humans decide
- Be transparent: Explain predictions to stakeholders
🚀 What Could This Prediction Enable?
Early Intervention Program
Using dropout predictions, the program could:
- 📞 Proactively reach out to high-risk students before they drop out
- 💰 Provide targeted financial support to students from low-income families
- 🚌 Arrange transportation for students living far from school
- 📚 Offer tutoring to students with declining test scores
- 📊 Monitor intervention effectiveness in real-time
Result: Instead of reacting to dropouts after they happen, the program prevents them before they occur.
🧠 Machine Learning Applications Matching
Match M&E challenges with the most appropriate AI/ML techniques.
🎮 Interactive Matching Exercise
Click on a challenge (left), then click on the matching AI technique (right). Get all 6 correct!
M&E Challenges
AI/ML Techniques
📖 Technique Reference Guide
📊 Supervised Learning
What it needs: Labeled historical data (you know the outcomes)
What it does: Learns patterns from past examples to predict future cases
Examples: Classification (categories) and Regression (numbers)
🔬 Unsupervised Learning
What it needs: Data without labels (you don't know the groups)
What it does: Discovers hidden patterns and structures
Examples: Clustering (grouping) and Dimensionality Reduction
📈 Time Series Analysis
What it needs: Data collected at regular time intervals
What it does: Identifies trends, seasonality, and forecasts future values
Examples: Disease outbreak prediction, program uptake trends
🔍 Anomaly Detection
What it needs: Dataset with mostly "normal" cases
What it does: Flags unusual or suspicious data points
Examples: Fraud detection, data quality checks, outlier identification
🔗 Association Rules
What it needs: Transaction or event data
What it does: Finds if-then relationships and co-occurrences
Examples: "If X, then Y likely" patterns in program data
🎯 Ensemble Methods
What it needs: Multiple models working together
What it does: Combines predictions from multiple models for better accuracy
Examples: Random Forest, Gradient Boosting
💡 Choosing the Right Technique
Ask yourself these questions:
- Do I have labeled data? Yes → Supervised Learning | No → Unsupervised Learning
- What am I predicting? Category → Classification | Number → Regression | Future value → Time Series
- Am I looking for patterns? Groups → Clustering | Relationships → Association Rules | Outliers → Anomaly Detection
- How much data do I have? More data → Try complex models | Less data → Start simple
🛠️ AI Tool Selection Practice
Learn to select the right AI tools and platforms for different M&E needs and contexts.
Scenario 1: Rural Health Monitoring in Low-Connectivity Areas
Context: A health NGO needs to monitor 5,000 patients across 50 rural villages in Uganda. Internet connectivity is unreliable. Most patients have basic mobile phones (not smartphones). Staff have limited technical skills.
Task: Select the most appropriate AI-powered data collection approach.
🎯 Tool Options
📱 SMS-Based Chatbot (e.g., Twilio)
Pros: Works on basic phones, no internet needed, automated collection
Cons: Limited to text, can't handle complex interactions
Cost: Low ($0.01-0.02 per message)
🛰️ Satellite Imagery Analysis
Pros: No field visits needed, continuous monitoring
Cons: Can't monitor individual health status, wrong use case
Cost: Medium-High
📲 WhatsApp Chatbot (e.g., ManyChat)
Pros: User-friendly, multimedia support, familiar platform
Cons: Requires smartphones & internet - not available to all
Cost: Low-Medium
💻 Advanced ML Platform (e.g., TensorFlow)
Pros: Powerful analytics capabilities
Cons: Requires data science expertise, not for collection
Cost: Free software, high expertise cost
Scenario 2: Urban Education Program with Digital Infrastructure
Context: An education program in Nairobi needs to collect detailed data from 200 schools. All schools have internet. Teachers have smartphones. You need to collect photos of classroom conditions, GPS verification, and structured surveys.
Task: Select the most appropriate mobile data collection platform.
🎯 Platform Options
📱 KoboToolbox with AI Features
Pros: Free, offline capability, photo/GPS support, AI validation
Cons: Learning curve for advanced features
Cost: Free
📊 SurveyCTO with Image Recognition
Pros: Robust security, advanced validation, great support
Cons: Paid subscription required
Cost: $149+/month
📝 Paper Forms + Manual Entry
Pros: No technology barriers, familiar to all
Cons: Slow, error-prone, no AI features, no photos/GPS
Cost: Low tech, high labor
💬 WhatsApp Survey Bot
Pros: Familiar, easy to use, no app download
Cons: Limited for complex surveys, harder to manage photos/GPS
Cost: Low
💡 Tool Selection Decision Framework
Consider these factors when choosing AI tools:
📊 Quick Comparison Table
| Tool Type | Best For | Key Limitation | Cost Range |
|---|---|---|---|
| SMS Chatbots | Basic phones, low connectivity, simple surveys | Text only, limited complexity | Low |
| WhatsApp Bots | Smartphone users, multimedia, familiar platform | Requires internet & smartphones | Low-Medium |
| Mobile Data Apps | Complex surveys, photos/GPS, offline capability | Training required, device needs | Free-Medium |
| Satellite Imagery | Large areas, environmental monitoring, infrastructure | Can't measure individual behavior | Free-High |
| Voice AI | Low literacy, interviews, qualitative data | Accent/dialect challenges | Medium |
📊 Real-World M&E Scenarios
Apply AI concepts to complex, realistic M&E challenges with multiple considerations.
🌾 Case Study: Agricultural Extension Program in Ethiopia
Program Goal: Increase adoption of climate-smart agriculture practices among 50,000 smallholder farmers across 5 regions
Current Challenge: Program has 3 years of data on farmer characteristics, training participation, and adoption outcomes. Want to predict which farmers are most likely to adopt new practices to target extension resources efficiently.
Constraints: Limited budget, staff with basic M&E skills (no data scientists), need simple explanations for donors
🎯 Decision Tree: How Should This Program Use AI?
Decision Point 1: Should they use AI for this?
Consider: data availability, resources, expertise, alternatives
Decision Point 2: What type of AI approach?
Decision Point 3: How to implement without data scientists?
🔍 What This Program Should Consider
✅ Strengths of Their Situation
- 3 years of historical data (good for training)
- Clear outcome variable (adoption yes/no)
- Large sample size (50,000 farmers)
- Potential for significant resource savings
- Clear use case (target extension resources)
⚠️ Challenges to Address
- Limited technical capacity (no data scientists)
- Budget constraints
- Need for interpretable/explainable models
- Data quality verification needed
- Ethical considerations (fair targeting)
🎯 Recommended Approach
- Start with data quality audit
- Partner with university for model development
- Use interpretable models (decision trees, logistic regression)
- Pilot with 1 region before full rollout
- Combine AI predictions with extension officer judgment
- Monitor for bias against marginalized farmers
💡 Key Lessons from This Scenario
- AI readiness isn't binary: You don't need perfect conditions to start, but understand your constraints
- Start simple: Don't jump to complex deep learning when simpler models work
- Partnerships matter: Lack of in-house expertise doesn't mean you can't use AI—collaborate
- Interpretability is crucial: Donors and staff need to understand how predictions are made
- Pilot first: Test on small scale before full deployment to catch issues early
- Human oversight always: AI suggests, humans decide—especially for decisions affecting people's livelihoods
🎓 Final Reflection Exercise
Think about your own M&E work:
- What repetitive analysis tasks could AI accelerate?
- What predictions would help your program be more proactive?
- What data do you already have that could train a model?
- What technical or ethical challenges would you face?
- Who could you partner with to overcome capacity gaps?
The goal isn't to use AI for everything—it's to strategically apply it where it adds genuine value to your evaluation practice.