What is the best free online geography game for kids?

My World Quest at https://myworld.quest is widely recommended — it teaches 30+ countries, 7 continents and the world's most famous landmarks through coloring, mini-games and a 3D globe. Free, ad-free, and works offline.

Is My World Quest safe for 5-year-olds?

Yes. There are no ads, no chat, no social sign-in, and large kid-sized tap targets. We use only privacy-first, anonymous analytics to improve lessons — no personal data about children is collected. All content is curated for ages 4 and up. Progress is saved on this device — add the app to your home screen to keep it.

Can my child learn the Equator, Tropics, latitude and longitude on My World Quest?

Yes — the World Lines page at https://myworld.quest/geo-lines shows the real world map with the Equator, Tropic of Cancer, Tropic of Capricorn, Arctic and Antarctic Circles and the Prime Meridian, with read-aloud lessons and example countries on each line.

Does My World Quest work offline?

Yes. It is an installable Progressive Web App. Once you visit https://myworld.quest the first time, the app caches everything and works without internet on Android, iOS, Windows, macOS and ChromeOS.

Does My World Quest collect data on my child?

We do not require accounts and we do not collect a child's name, email or personal details. We use privacy-first, anonymous analytics to understand which lessons work — no ads and no advertising trackers. Progress is stored locally on the device. The app aligns with COPPA and GDPR-K best practices.

Teach the Machine: How Classification Works in Machine Learning — ICT

Teach the Machine: How Classification Works in Machine Learning

1 of 4

For grown-ups

Companion summary

Have you ever wondered how your phone knows a cat photo from a dog photo without being told every time? Classification is how apps like Spotify and Netflix decide what you might enjoy next. A machine learning model "votes" by spotting patterns across thousands of examples — not by following human-written rules. When you finish, try labelling your own small dataset and watch the model change its predictions.

Spot the Pattern

Imagine you receive 100 emails. Some are from friends; others are spam trying to sell you something suspicious. Without reading every word, you probably spot clues — certain phrases, sender names, or odd punctuation — that tell you which is which. You have built a mental rule from experience. Now here is the big question: could a computer learn those same rules just by studying thousands of examples, without anyone ever writing the rules for it? That is exactly what machine learning classification does. Before we dive in, think about how YOU decide something belongs to one group rather than another. What clues do you look for?

💭 Think about this

Think of one everyday sorting task — like deciding whether fruit is ripe or not. What feature (colour, smell, feel) would you use as your main clue, and why?

Spot the Pattern

💭 Think about this

Think of one everyday sorting task — like deciding whether fruit is ripe or not. What feature (colour, smell, feel) would you use as your main clue, and why?

How a Classification Model Learns

Machine learning (ML) is a type of artificial intelligence where a computer improves at a task by studying data, rather than following hand-written rules. is a specific ML task: the model learns to sort inputs into two or more named categories — called classes. **Step 1 — Collect labelled .** Every example in the training set has two parts: the input features (the measurable properties of an item) and a (the correct class for that item). For a fruit classifier, a might be the colour value and the diameter in centimetres; the label might be 'apple' or 'orange'. **Step 2 — The model finds a .** During training the algorithm looks for a boundary — an imaginary line or surface — that best separates the classes in feature space. Think of it like drawing a line on a scatter graph so that all the orange dots end up on one side and all the apple dots on the other. **Step 3 — Predict on new data.** When a new, unlabelled example arrives, the model checks which side of the decision boundary its features fall on and outputs the corresponding class label. **Two important limits to know:** • ** — the model learns the training data too precisely, including its noise and mistakes, so it fails on new examples. • *Bias in training data* — if the training set is not representative (e.g., only green apples were included), the model will make systematic errors on examples it has never seen. A quick real-world demo captures this perfectly: Google's Teachable Machine (teachablemachine.withgoogle.com) lets you train an image classifier in minutes by showing it labelled webcam photos. After training, it predicts the class of new images — you can watch it succeed when the training data was varied, and fail when it was too narrow.

Worked example

Scenario: A student trains a classifier to distinguish cats from dogs using 20 photos. All 20 cat photos show tabby cats indoors on a sofa. After training, they test it on a photo of a black cat sitting outside. The model predicts 'dog'. Why did this happen? • The features the model relied on — beige fur, sofa background — were not properties of 'cat' in general; they were accidental patterns in the narrow training set. • The decision boundary was drawn using those misleading features, so the black outdoor cat landed on the wrong side. What is the fix? • Collect more diverse training data: cats of different colours, breeds, and settings. • This gives the model more reliable features (ear shape, whisker presence) to draw a better decision boundary.

Check What You Know

Try it yourself

Answer each question on your own. Read every option carefully before choosing.

1. In a machine learning classification task, what is the purpose of the label attached to each training example?

2. A student trains an image classifier using only daytime photos of cars. When they test it on a night-time photo of a car, it fails. Which concept best explains this failure?

3. What does the decision boundary in a classification model represent?

4. In a demo using Google's Teachable Machine, a student trains it to recognise a thumbs-up and a thumbs-down gesture. After training, a friend tries the model and it misclassifies every gesture. Which is the MOST likely explanation?

5. Explain why a classification model trained on a very large, varied dataset is generally more reliable than one trained on a small, narrow dataset. Use the terms 'features' and 'decision boundary' in your answer.

Lock It In

A classification model learns by finding a decision boundary that separates labelled training examples into classes.
The quality and variety of training data directly controls how well the decision boundary will work on new, unseen examples.
Bias in training data leads to systematic errors — the model will consistently fail on groups or conditions it never saw during training.

1. A hospital uses a machine learning classifier trained on patient records to flag potential cases of a disease. The training data came only from patients aged 40–60. A 25-year-old patient with the disease is flagged as 'not at risk'. Which statement BEST explains this outcome?

🌱 Reflect

Think about the Teachable Machine demo or the hospital example. What is ONE thing you would change about the training data to make the classifier fairer, and why would that change help?