Machine learning, explained with real numbers
Short lessons in plain language — no math background needed. Every idea is taught through a worked example you can follow by hand, and every term links to a bilingual glossary aligned with the official SDAIA AI glossary.
The lessons
Six short reads, in order — or jump straight to the one you need.
What is a model & how training works
Where the "model" actually comes from: training, testing, and why holding data back matters.
Read the lessonRegression by worked example
Predict a number — five houses, one formula, and every error metric worked out by hand.
Read the lessonClassification by worked example
Predict a category — ten customers, a hand-filled confusion matrix, and what accuracy hides.
Read the lessonForecasting: sales next month
Predict the future from the past — lags, horizon, and why far-out forecasts drift.
Read the lessonUnderstanding your metrics
R², MAE, precision, recall and friends — what each score means and when to trust it.
Read the lessonReading your dataset report
The thirty-ish automatic checks Bayanii runs on your data, and what each finding asks of you.
Read the lessonBefore you train
Two things worth knowing before your first model: the mistakes that fool most beginners, and when a model is the wrong tool.
Common Mistakes
Overfitting
When your model learns the training data too well—memorizing every detail instead of learning general patterns. It performs great on training data but poorly on new data.
Signs: High accuracy on training data but low accuracy on test data. Solution: Use more data, simplify your model, or use regularization techniques.
Bad Data
Garbage in, garbage out. If your data has errors, missing values, or doesn't represent the real world, your model will learn the wrong patterns.
Common issues: Missing values, typos, inconsistent formats, outdated information. Solution: Clean your data before training—remove duplicates, fill missing values, and check for errors.
Data Leakage
When information from the future or the answer itself accidentally gets into your training data. Your model seems perfect, but it's cheating.
Example: Predicting customer churn but including "days since last purchase" when that information wouldn't be available at prediction time. Solution: Only use features that will be available when making real predictions.
When NOT to use AI
Machine learning isn't always the right solution. Here are situations where you should consider alternatives:
- Simple rules work better: If a simple if-then rule solves your problem (like "if age > 18, then eligible"), use that instead. It's faster, cheaper, and easier to understand.
- Not enough data: Machine learning needs examples to learn from. If you have very few data points (like 10-20), traditional statistics or expert knowledge might be better.
- No clear pattern: If your data is completely random with no underlying patterns, no model can learn anything useful.
- Explainability is critical: In regulated industries (healthcare, finance), you may need to explain every decision. Some AI models are "black boxes" that are hard to explain.
- Real-time requirements: If you need instant results and can't wait for model training, a pre-built solution or rule-based system might be faster.
- Cost outweighs benefit: If building and maintaining a model costs more than the value it provides, stick with simpler solutions.
Remember: AI is a tool, not a magic solution. Use it when it adds value, not just because it's trendy.
Every term, in both languages
The glossary defines every metric and concept the lessons use — in plain English and Arabic, with a tiny worked example for each. Arabic terms follow the official SDAIA AI glossary.