Abstract

Early identification of learning disabilities is critical for effective intervention and support. This paper proposes a novel deep learning framework that analyzes multimodal student interaction data—including handwriting patterns, reading behaviors, eye-tracking metrics, and response latencies—to detect early signs of dyslexia, dyscalculia, and attention deficit disorders. The proposed CNN-LSTM hybrid model was trained on a dataset of 2,400 primary school students across three countries and achieved a classification accuracy of 94.3% for dyslexia detection and 91.7% for dyscalculia identification. The system provides teachers with interpretable risk assessments and personalized intervention recommendations. A pilot deployment in 12 schools demonstrated that early AI-assisted detection led to 40% faster referral to specialist services compared to traditional screening methods. Ethical considerations regarding student data privacy and the role of AI as a complementary tool to professional diagnosis are thoroughly discussed.