Different problems need different models. This week you meet the five workhorses of machine learning and drive each one yourself: linear regression, logistic regression, k-means clustering, decision trees, and time series forecasting. Every lab uses Caribbean data, because that is the data you will build with.
How machines learn from data, and which model fits which job: predicting a number, predicting a yes/no, finding groups, making rule-based decisions, and forecasting the future. Work through the five labs on this page before Friday.
Explore how AI reads satellite and earth data, the same imagery used to watch storms form in the Atlantic and track sargassum heading to our beaches.
Bring a real problem from home. Together we decide which of the five models fits it, and start building the solution.
Linear regression predicts a number by drawing the best straight line through data. Here, each blue dot is one day at a Half Way Tree sky juice stand: how hot it was, and how many cups sold. Hotter day, more cups. The model's job is to find the line that fits that pattern best.
Sometimes you do not want a number, you want a yes or no with a probability. Logistic regression fits an S-shaped curve that turns any input into a probability between 0% and 100%. Each dot below is a past storm day: rainfall along the bottom, and whether the gully flooded (top row) or not (bottom row).
The first two models learned from labelled examples. K-means is different: it finds groups nobody labelled. Each dot is a beach visitor in Montego Bay: hours spent on the beach, and money spent. Nobody told the computer who is who. Watch it discover the groups itself: this is called unsupervised learning.
A decision tree is a model that learned which questions to ask, and in which order. Each split was chosen from data because it best separates the outcomes. You are the principal this morning: walk the tree and let it guide your call.
A time series model learns from the past of one measurement to forecast its future. This chart shows two years of monthly visitor arrivals to Jamaica (in thousands). See the repeating shape? High season in winter, a summer bump, a September to October dip. That repeating shape is called seasonality, and it is exactly what the model learns.
Whichever of the five models you choose, it only learns from the data you feed it. A "feature" is one input column. You are building a model to predict flooding in Portmore: choose wisely.
1. Model Matchmaker (Tuesday). Write down five problems from your own life or your community, one for each model type. Example: "predict my phone data usage next month" is a time series forecast; "will the bus come before 7:15, yes or no" is logistic regression territory. Bring your five to Friday's session.
2. Problem Hunt (Thursday). Interview one parent or guardian. Ask them: "What is the most annoying, repetitive problem in your week?" Write it down in one sentence, then decide which of the five models could attack it.
3. Flood model rematch (weekend). Go back to the AI Quests flood mission and try to beat your Week 1 accuracy. You now know what the model is doing under the hood. Submit your new score at the same link.
Replay the Flood Prediction quest → Submit your improved score →