Machine Learning — How Computers Learn from Data

What is machine learning?

Machine learning (ML) is a subset of AI where systems improve from experience (data) without being explicitly programmed for every rule.

Instead of writing "if subject contains 'winner' then spam," you show thousands of labeled emails and the model learns distinguishing patterns.

Types of machine learning

• check_circleSupervised learning — labeled examples (input → known output). Used for prediction, classification, regression.
• check_circleUnsupervised learning — no labels; find structure (customer segments, anomalies).
• check_circleSemi-supervised — small labeled set + large unlabeled set.
• check_circleReinforcement learning — agent learns by trial and reward (robotics, game AI, some recommendation systems).

Common algorithms (conceptual)

Deep learning (Step 6) dominates images, text, and speech — but classical ML still powers many enterprise dashboards and credit models.

• check_circleLinear & logistic regression — simple, interpretable baselines for numeric and yes/no predictions.
• check_circleDecision trees & random forests — rule-based splits; strong on tabular business data.
• check_circleSupport Vector Machines (SVM) — effective on smaller structured datasets.
• check_circleGradient boosting (XGBoost, LightGBM) — often wins Kaggle tabular competitions.
• check_circleK-means clustering — group similar customers or products.

The ML workflow

• check_circle1. Problem definition — what decision are we automating?
• check_circle2. Data collection & cleaning — often 80% of the work.
• check_circle3. Feature engineering — turning raw fields into useful inputs.
• check_circle4. Model selection & training — pick algorithm, train, tune hyperparameters.
• check_circle5. Evaluation — metrics on hold-out data.
• check_circle6. Deployment — integrate into app or batch pipeline.
• check_circle7. Monitoring — models drift as the world changes; retrain periodically.

Overfitting vs underfitting

Overfitting: model memorizes training data, fails on new data. Like a student who memorizes answers but can't solve new problems.

Underfitting: model too simple to capture patterns. Fix by better features, more data, or richer models — balanced with validation metrics.

ML in business today

Rwandan fintech and agritech startups increasingly use these techniques — often starting with spreadsheets before graduating to custom pipelines.

• check_circleCredit scoring and fraud detection in banking
• check_circleDemand forecasting in retail and logistics
• check_circleChurn prediction for telcos and SaaS
• check_circleRecommendation engines (Netflix, e-commerce)
• check_circlePredictive maintenance in manufacturing