This research introduces an adaptive e-learning framework powered by a multi-label classification model using Bidirectional Long Short-Term Memory (BiLSTM) networks, designed to support children with diverse learning and developmental disabilities . The system achieves high precision, with an average F1 score of 0.94 across diagnostic categories, and demonstrates comparative advantages over existing tools by incorporating gamified and multi-sensory modules. By addressing critical gaps in inclusive education , the framework identifies multiple disabilities such as cognitive, auditory, and visual impairments . The architecture employs robust preprocessing, advanced feature engineering, and optimization techniques, ensuring scalability and high accuracy . This work contributes significantly to advancing inclusive education by combining state-of-the-art machine learning with practical accessibility solutions . Future directions include enhancing multilingual support, gamification, and real-time tracking features to improve engagement and accessibility in diverse educational contexts.