Towards the development of a multi-axial classification of people with learning difficulties.

The team pulled 30 years of records on people with learning disabilities. They fed the data into a computer and let it sort people into groups.

The program made ten clusters based on skills, behaviors, and health needs. No expert labels were used — just the numbers.

The ten-cluster picture looked sensible to clinicians. Each group had clear strengths and needs.

Still, the authors said the map needs more testing before anyone uses it for placement or funding.

Mélin et al. (2023) took the next step. Their data-driven clusters of preschoolers with mixed delays predicted who would gain the most adaptive skills two years later.

Préfontaine et al. (2024) and Marsack-Topolewski et al. (2025) swapped the old clustering for machine-learning models. Both teams forecast autism intervention success with 97-a large share accuracy, showing the idea scales when you add modern code.

Bailey et al. (2021) used the same math trick on QABF data. Their neural net guessed behavioral function better than standard rules, proving the approach works inside single-case work too.

You already write detailed intake reports. Run a quick cluster or ML script on those numbers and you get a forecast of which clients will need heavier staff ratios or will rocket once ABA starts. The 1994 map is rough, but the follow-up papers give you free code that sharpens it. Try one this month — it takes ten minutes in Excel or R and could save you weeks of trial-and-error programming later.