AI and programming classes in Ellesmere Port

The learner collects Census 2021 data from the Nomis API for all 1,179 output areas in Cheshire West and Chester. Of 155,125 households, 25,879 have no car or van, and the goal is to predict each area's no-car share. Inputs come in two kinds: where the area is, taken from the ONS population-weighted centre point, and what it is like, measured by population density and the share of one-person households. A random forest model is trained and tested with five-fold cross-validation, first splitting the areas at random, then holding out whole neighbourhoods at once, using the 47 larger areas (MSOAs) the ONS groups them into.

Before any model, the learner checks how alike neighbours are. Moran's I, a standard measure of spatial autocorrelation that runs from about minus one to one, comes out at 0.558 for the no-car share when each area is compared with its eight nearest neighbours. That is strong: knowing the neighbours tells you a lot.

With a random split, the location-only model looks respectable, cutting the error from 10.73 to 6.87. Hold out whole neighbourhoods and it jumps to 8.86, nearly 30% worse, because it can no longer look up the answer from the house next door. The model built on density and household mix barely changes, 6.39 against 6.48: it had learned something about areas in general, not about particular places. The combined model still leads under both tests, but its lead narrows once the leak is closed. A score is only as honest as the split behind it.

A teenager in Ellesmere Port who can question the test as well as the score will not be fooled by a dashboard, and coding is how that habit forms. The longer argument is in why teenagers should still learn to code in 2026.