Exploring and Identifying Key Factors in Predicting Dyslexia in Children: Advanced Machine Learning Algorithms From Screening to Diagnosis
Fuente:
Wiley Open Access
Saved in:
| Main Author: | |
|---|---|
| Format: | Artículo Open Access |
| Published: |
Wiley
2025
|
| Subjects: | |
| Online Access: | |
| Tags: |
Add Tag
No Tags, Be the first to tag this record!
|
| _version_ | 1867005839636692992 |
|---|---|
| author | Abdullah Alrubaian |
| author_facet | Abdullah Alrubaian Abdullah Alrubaian |
| collection | Wiley Open Access |
| contents | Exploring and Identifying Key Factors in Predicting Dyslexia in Children: Advanced Machine Learning Algorithms From Screening to Diagnosis Abdullah Alrubaian Clinical Psychology & Psychotherapy ABSTRACT Introduction The current study aimed to develop and validate a machine learning (ML)–based predictive models for early dyslexia detection in children by integrating neurocognitive, linguistic and behavioural predictors. Method A cross‐sectional study was conducted with 300 Saudi Arabian children (150 children with dyslexia, 150 controls) aged 6–12 years and their parents. Participants underwent assessments for attention, phonological awareness, rapid automatised naming (RAN), cognitive flexibility and other predictors. Four ML models—logistic regression, random forest, XGBoost and an ensemble—were trained and evaluated using performance metrics (AUC, sensitivity, specificity). Recursive feature elimination (RFE) identified key predictors. Results The RFE (15‐fold cross‐validation) identified attention, RAN, early language delay, phonological awareness and cognitive flexibility as the top five predictors of dyslexia. The ML models demonstrated high diagnostic accuracy for dyslexia detection. Logistic regression achieved superior performance with an area under the curve (AUC) of 0.95 (95% CI: 0.92–0.98), sensitivity of 97%, specificity of 91% and overall accuracy of 94%. Random forest and XGBoost yielded slightly lower but robust AUCs (0.91 and 0.93, respectively), with balanced sensitivity (95%) and specificity (91%). The ensemble model harmonised algorithmic strengths, retaining an AUC of 0.93 while prioritising interpretability through weighted contributions from XGBoost (40%), random forest (30%) and logistic regression (30%). Conclusion This study demonstrated the transformative potential of ML in dyslexia diagnostics. By systematically prioritising phonological awareness, RAN and attention deficits, ML models offer a scalable, objective framework for early identification. These tools could alleviate reliance on subjective assessments, enabling timely interventions to mitigate dyslexia's long‐term impacts. 10.1002/cpp.70077 http://onlinelibrary.wiley.com/termsAndConditions#vor |
| doi_str_mv | 10.1002/cpp.70077 |
| format | Artículo Open Access |
| id | wiley_oa_10_1002_cpp_70077 |
| institution | Wiley Open Access |
| license_str_mv | http://onlinelibrary.wiley.com/termsAndConditions#vor |
| publishDate | 2025 |
| publisher | Wiley |
| record_format | wiley_oa |
| spellingShingle | Exploring and Identifying Key Factors in Predicting Dyslexia in Children: Advanced Machine Learning Algorithms From Screening to Diagnosis Abdullah Alrubaian Clinical Psychology & Psychotherapy Exploring and Identifying Key Factors in Predicting Dyslexia in Children: Advanced Machine Learning Algorithms From Screening to Diagnosis Abdullah Alrubaian Clinical Psychology & Psychotherapy ABSTRACT Introduction The current study aimed to develop and validate a machine learning (ML)–based predictive models for early dyslexia detection in children by integrating neurocognitive, linguistic and behavioural predictors. Method A cross‐sectional study was conducted with 300 Saudi Arabian children (150 children with dyslexia, 150 controls) aged 6–12 years and their parents. Participants underwent assessments for attention, phonological awareness, rapid automatised naming (RAN), cognitive flexibility and other predictors. Four ML models—logistic regression, random forest, XGBoost and an ensemble—were trained and evaluated using performance metrics (AUC, sensitivity, specificity). Recursive feature elimination (RFE) identified key predictors. Results The RFE (15‐fold cross‐validation) identified attention, RAN, early language delay, phonological awareness and cognitive flexibility as the top five predictors of dyslexia. The ML models demonstrated high diagnostic accuracy for dyslexia detection. Logistic regression achieved superior performance with an area under the curve (AUC) of 0.95 (95% CI: 0.92–0.98), sensitivity of 97%, specificity of 91% and overall accuracy of 94%. Random forest and XGBoost yielded slightly lower but robust AUCs (0.91 and 0.93, respectively), with balanced sensitivity (95%) and specificity (91%). The ensemble model harmonised algorithmic strengths, retaining an AUC of 0.93 while prioritising interpretability through weighted contributions from XGBoost (40%), random forest (30%) and logistic regression (30%). Conclusion This study demonstrated the transformative potential of ML in dyslexia diagnostics. By systematically prioritising phonological awareness, RAN and attention deficits, ML models offer a scalable, objective framework for early identification. These tools could alleviate reliance on subjective assessments, enabling timely interventions to mitigate dyslexia's long‐term impacts. 10.1002/cpp.70077 http://onlinelibrary.wiley.com/termsAndConditions#vor |
| title | Exploring and Identifying Key Factors in Predicting Dyslexia in Children: Advanced Machine Learning Algorithms From Screening to Diagnosis |
| topic | Clinical Psychology & Psychotherapy |
| url | https://onlinelibrary.wiley.com/doi/10.1002/cpp.70077 |