A 3-Tier machine Learning Framework for Early Detection of Learning Difficulties in Basic setting

Early identification of learning difficulties (LDs) in elementary education is crucial for timely intervention, however, conventional approaches relying on subjective teacher assessments often result in delayed or inequitable support, particularly affecting underserved populations. This study proposes and evaluates a scalable machine learning (ML) based framework to support the early identification of learning difficulties (LDs) among upper primary school learners in Ghana’s basic education system. Data were collected from 1,124 learners across public schools over two academic years (2021–2023), encompassing psychometric, behavioural, and demographic variables. Using validated instruments such as the Learning Difficulties Checklist (LDC), Conners’ Teacher Rating Scale (CTRS), and WRAT subtests, learners were categorized into low, moderate, and high LD risk groups based on the composite score thresholds established through statistical analysis and expert input. A hybrid stacking ensemble model, which combines Random Forest, XGBoost, Support Vector Machine, and Logistic Regression outperformed individual models, achieving 95% accuracy, a macro – average F1 score of 93% and an ROC-AUC of 99.5%. Key predictors included working memory, rapid naming, math attitude, ADHD symptoms, and vocabulary. Construct reliability and validity were confirmed using Cronbach’s alpha and ANOVA, with stratified 10 – fold cross validation ensuring model robustness. The proposed framework is designed to guide teachers’ decision making and facilitate the early diagnosis of learning difficulties, ensuring inclusive education. Ethical considerations (data anonymization and ethical consent were sought and approved from designated institutions and participants respectively) were considered throughout the study to support responsible research conduct and future application. The findings underscore the potential of ML-driven tools to enhance special education by enabling proactive data-informed instructional planning in under -resourced contexts.