Quest 3 of 15
Machine Learning Basics
Understand how models learn from data — labels, train/test splits, overfitting, and metrics — using interactive diagrams and concept labs. No coding is required on this platform, but you will learn the reasoning needed to judge simple model claims. For example, you will compare false positives and false negatives in a spam filter and disease-screening scenario to decide which mistake matters more.
Start here
Machine learning sounds advanced, but the core idea is simple: show examples, let the computer find patterns, then test whether it works on new examples it has never seen.
Big idea
Before a model is used in the real world, builders split data into training and testing sets, watch for overfitting, and choose metrics that match what actually matters when the model is wrong.
Machine learning lifecycle
Collect examples with correct answers attached.
Learn one idea at a time
Read, explore, then mark each idea when you can explain it.
Idea 1 of 7
Data is the raw material of machine learning: a collection of recorded examples such as rows in a spreadsheet, photographs, emails, or sensor readings. Each example describes one case — one loan application, one leaf photo, one message. The quality of this collection sets a ceiling on everything that follows. If the examples are full of errors, missing important situations, or collected from only one group of people, no algorithm, however sophisticated, can recover the information that was never captured. Practitioners summarise this as “garbage in, garbage out,” and it is the reason experienced teams spend more time on data than on models.
Choose a deep dive
Open the topics you want to explore. The detail stays folded until you need it.
Deep dive 1Testing is like a new exam
A model sees the training set while it is learning, so success there does not prove it will work in real life. The test set contains examples it did not see before. If performance drops sharply on that set, the model may have memorised instead of learned a useful rule.
Deep dive 2Not all errors have the same cost
A model that filters spam can make occasional mistakes with low harm. A model helping to identify crop disease or prioritise clinic visits needs more careful testing because a missed case can matter greatly. The right measure depends on who is affected by each type of error.
Deep dive 3Case study: the model that aced its test and failed in the field
A team builds a model to spot diseased cassava leaves from photos and reports 96% accuracy. In the field, farmers find it nearly useless. The investigation reveals the training photos were all taken by one researcher: same camera, same angle, same time of day, leaves picked and placed on a white sheet. Farmers photograph leaves on the plant, in shade, on cheap phones. The model had learned the researcher's photography style, not the disease. The fix was collecting new photos in real conditions from many farms. The score was never wrong — the test simply did not represent reality, which is the most common way evaluation misleads people.
Deep dive 4Reading a confusion matrix without fear
A confusion matrix is just a two-by-two count: cases the model correctly flagged, cases it correctly cleared, false alarms, and misses. From it come the two questions that matter in practice. Of everything the model flagged, how much was real (precision)? Of everything real, how much did the model catch (recall)? A disease screener may accept more false alarms to catch every real case; a spam filter may do the opposite because deleting a real message is worse than seeing one extra spam. Once you can ask 'which error is worse here?', you can evaluate any classifier sensibly — no mathematics degree required.
Deep dive 5Train then test
Hold back a test set like an unseen exam.
One-minute challenge
Connect this lesson to real life
Name one situation where this idea could help, and one thing a person should still check.
Explore a real-world example
Use the arrows to connect the idea to a visible situation.
Photo example
Example: labels matter
Wrong labels teach wrong patterns.
Key terms
Tap a term to flip and read the definition.
Optional further learningFree textbooks and trusted online resources
These sources informed the course structure. Use them to revisit a concept or study it in more depth.
Ready check
Tick each idea only when you could explain it without looking back.
Ready for practice? Use the match and quiz activities to connect these terms to real model behaviour before moving on.
Extra context (audience, logistics, curriculum notes)
Built for: Concept-first track on this site. Learners who want Python notebooks should use an external sandbox alongside these modules.
Formats: Interactive pipeline diagrams · Concept matching and scenarios · Optional external notebook (not hosted here)
Module 3 — ML concepts without code; optional coding path noted for external sandbox learners.
Next up
Ready for the next part?
When you've finished the reading, inline exercises, and knowledge check for this part, check the box to continue.