AI and programming classes in East Kilbride

The learner downloads OpenStreetMap data covering East Kilbride in twelve tiles and joins the road pieces tagged as roundabouts into complete rings: 92 of them. For comparison, it collects the 111 other road ways that close on themselves, mostly residential and service roads such as loops and turning circles. For every ring it computes circularity from the area and perimeter, and a second feature: how many other road pieces (OpenStreetMap ways) touch it. The program then tries simple rules and counts how many roundabouts it finds (recall) and how many of its claims are right (precision).

Circularity alone finds nearly every roundabout but raises 62 false alarms, because many of the other loops are round as well: their median circularity is 0.91, not far below the roundabouts' 0.992. A turning circle at the end of a close is a neat circle too. What gives it away is that it touches a single other road piece, as 60 of the 111 other loops do, whereas a roundabout is where roads meet. Adding that second feature cuts the false alarms to 15 while losing eight genuine roundabouts, those touched by fewer than three road pieces. Choosing and combining clues like this is called feature engineering, and it often matters more than the choice of model.

The remaining 15 false alarms deserve a look rather than a shrug. Some may be loop roads; some could be roundabouts that nobody has tagged yet. A model scored against map labels can only be as right as the labels.

Choosing what a model should look at keeps East Kilbride teenagers in charge of the AI they build, and that is a strong reason to learn to code in 2026. The longer argument is in why teenagers should still learn to code in 2026.