Pubudu — Deep-Learning Screening for Dyslexia, Dysgraphia and Dyscalculia in Sri Lanka

Researchers at the Sri Lanka Institute of Information Technology (SLIIT) in Malabe, near Colombo, developed Pubudu, a mobile application using deep neural networks (for dyslexia and dysgraphia screening) and other machine-learning models (for dyscalculia) to screen children aged 6-7 for these learning disabilities in local languages, following clinical screening protocols.

To screen young children for dyslexia, dysgraphia and dyscalculia, followed by a gamified intervention module after a positive screening result.

The team tested the system against 50 children with diagnosed learning disabilities and 50 without, winning the AI Innovator category at SLIIT Codefest 2019 and publishing results in the peer-reviewed IEEE ICIIS 2019 conference proceedings.

Reported accuracy was 88% for letter dysgraphia, 99% for numeric dysgraphia, 90% for dyscalculia, and 58% for dyslexia — the last result sitting close to chance level for a binary screening task, reported without smoothing over the weak result.

No evidence was found of Pubudu being deployed at scale in Sri Lankan schools since its 2019-2020 publication and award, nor of any follow-up study improving dyslexia-detection accuracy or reporting real-world usage numbers; it should be read as a validated research prototype rather than a proven, widely-deployed classroom tool.