Randomized trial evaluates a cloud-native screening system for early literacy in first-graders, suggesting improved accessibility and timely identification.
Key Points
The aim is to develop a scalable, cloud-based early literacy screening system that combines machine learning with human validation to enhance student assessments.
Developed K-KOBUKI, a cloud-based workflow for early literacy assessment using automated features and human verification.
Evaluated using data from 195 first-grade students with repetitive stratified cross-validation.
Incorporated machine learning classification models to identify at-risk students based on oral reading performance.
Multiple classification models achieved stable recall (≈0.85) under class imbalance conditions.
Psychometric-informed feature refinement improved precision without compromising recall.
Explainable AI analyses revealed strong contributions of word reading and reading fluency to model decisions.