Proposed Hybrid Intelligent Model Based on Artificial Neural Networks and Decision Trees for Early Detection of Learning Disabilities: Design and Performance

Abstract

Early identification of learning difficulties is an important challenge in education because timely screening can support appropriate educational intervention. Machine-learning methods can process multidimensional information, but individual approaches involve trade-offs between predictive performance and interpretability. This study proposes a sequential hybrid model that combines a C4.5 decision tree with a multilayer perceptron (MLP). In the first stage, the decision tree identifies informative features and generates human-readable IF–THEN rules; in the second stage, the selected features are provided to an MLP for three-class classification. The model was evaluated on a synthetic dataset of 1,200 children aged 5–7 years, containing 18 simulated demographic, cognitive, linguistic/academic, and behavioral features and three target classes: Typical, At-Risk Dyslexia, and At-Risk Dyscalculia. The hybrid model achieved 96.8% accuracy and a macro F1-score of 0.97 on the reported evaluation set, compared with 94.2% and 0.94 for the standalone MLP and 88.6% and 0.88 for the standalone decision tree. Because the experiments used synthetic data exclusively, these findings constitute a methodological proof of concept rather than evidence of clinical or educational diagnostic validity. Validation using independently collected real-world datasets is required before practical deployment.