Implementation of the Random Forest Algorithm in an Educational Game for Early Screening of Dyslexia Risk in Preschool Children

Dyslexia is a specific learning disorder affecting an estimated 5 17% of children, and identifying risk during the preschool years enables timely support. In Indonesian early childhood education (PAUD), however, risk identification still relies largely on teachers’ subjective observation, and no available tool combines the main pre-reading predictors with automated, offline analysis. This study aimed to develop and evaluate a prototype Machine Learning-Based Dyslexia Screening Game for children aged 4 6 years. Using a research and development design guided by CRISP-DM, four Indonesian-language mini-games were built in Unity Engine to measure phonological awareness, working memory, rapid automatized naming, and visual recognition without requiring reading ability. Eleven statistical features were extracted from logged interaction data, and a Random Forest classifier assigning low, moderate, or high risk was trained and tested (80:20 split) on a public reference dataset of 3,644 native Spanish-speaking participants aged 7 17 years (392 with a professional dyslexia diagnosis [10.7%]), mapped into derived low-, moderate-, and high-risk levels. The model was converted to ONNX via hummingbird-ml and executed on-device through Unity Barracuda. Separately, functional testing and a field usability trial with 15 children and 3 educators were conducted at PAUD KB Al Fitriyah. On the reference test set, the model achieved 87.3% accuracy, 84.6% macro precision, 82.1% macro recall, and a macro F1-score of 0.83, although High Risk recall was lower (78.0%). In the usability trial, 14 of 15 children (93.3%) played independently, with a mean completion time of 4.2 minutes. The findings demonstrate the technical feasibility of offline, game-based risk screening; however, the classification accuracy has not been validated among Indonesian preschool children, and the small single-site usability sample limits generalization. The prototype should therefore be regarded as a supplementary pre-screening aid, not a diagnostic instrument, pending local and clinical validation.

Authors

Institutions

Publication Details

Journal
Journal of Mathematics Instruction Social Research and Opinion
Published
2026-11-21
DOI
https://doi.org/10.58421/misro.v5i4.2170
Primary Topic
Reading and Literacy Development
Type
article
Field-Weighted Citation Impact
0.00
Controls
|||
ALL TIME
JAN
FEB
MAR
APR
MAY
JUN
JUL
AUG
SEP
OCT
NOV
article

Implementation of the Random Forest Algorithm in an Educational Game for Early Screening of Dyslexia Risk in Preschool Children

Luqman Rohmad Maghribi, Septa
Journal of Mathematics Instruction Social Research and Opinion
Reading and Literacy Development
article

Implementation of the Random Forest Algorithm in an Educational Game for Early Screening of Dyslexia Risk in Preschool Children

Luqman Rohmad Maghribi, Septa
article en

Abstract

Dyslexia is a specific learning disorder affecting an estimated 5 17% of children, and identifying risk during the preschool years enables timely support. In Indonesian early childhood education (PAUD), however, risk identification still relies largely on teachers’ subjective observation, and no available tool combines the main pre-reading predictors with automated, offline analysis. This study aimed to develop and evaluate a prototype Machine Learning-Based Dyslexia Screening Game for children aged 4 6 years. Using a research and development design guided by CRISP-DM, four Indonesian-language mini-games were built in Unity Engine to measure phonological awareness, working memory, rapid automatized naming, and visual recognition without requiring reading ability. Eleven statistical features were extracted from logged interaction data, and a Random Forest classifier assigning low, moderate, or high risk was trained and tested (80:20 split) on a public reference dataset of 3,644 native Spanish-speaking participants aged 7 17 years (392 with a professional dyslexia diagnosis [10.7%]), mapped into derived low-, moderate-, and high-risk levels. The model was converted to ONNX via hummingbird-ml and executed on-device through Unity Barracuda. Separately, functional testing and a field usability trial with 15 children and 3 educators were conducted at PAUD KB Al Fitriyah. On the reference test set, the model achieved 87.3% accuracy, 84.6% macro precision, 82.1% macro recall, and a macro F1-score of 0.83, although High Risk recall was lower (78.0%). In the usability trial, 14 of 15 children (93.3%) played independently, with a mean completion time of 4.2 minutes. The findings demonstrate the technical feasibility of offline, game-based risk screening; however, the classification accuracy has not been validated among Indonesian preschool children, and the small single-site usability sample limits generalization. The prototype should therefore be regarded as a supplementary pre-screening aid, not a diagnostic instrument, pending local and clinical validation.

Journal of Mathematics Instruction Social Research and OpinionVol. 5(4)
Universitas Pamulang (ID)
Reading and Literacy Development
AI Navigator

Ask Laika to Summarize, Analyze, and Connect papers live on the map.

Summarize Papers & Methodologies

Extract key findings, datasets, and comparative methods across publications.

Benchmark Rankings & Visual Analytics

Rank top research institutions, authors, funders, topics, and journals by Field-Weighted Citation Impact (FWCI) and paper volume with instant charts.

Connect Distant Disciplines

Bridge topological clusters on the map to find hidden collaborative intersections.

Laika access coming soon