An Explainable Deep Learning Framework for Early Identification of Learning Challenges in Students with Attention Deficit Hyperactivity Disorder

Abstract

There are some students who experience learning challenges that may manifest as Attention Deficit Hyperactivity Disorder (ADHD) that impact on their attention, cognitive processing, learning outcomes and engagement in class and in the learning environment. The traditional assessment methods are mainly based on behavioral observation and standardized assessment, and are not scalable and interpretable. However, these limitations can be overcome by the proposed Explainable Deep Learning Framework for early identification of learning difficulties in ADHD students based on multimodal educational and behavioral data. The framework combines a Convolutional Neural Network (CNN) for feature extraction, a Bidirectional Long Short-Term Memory (BiLSTM) network for modeling temporal learning patterns, and an attention mechanism to focus on the most relevant features affecting student performance. Explainability is embedded in the analysis, using SHAP-based analysis, which allows for an understandable interpretation of the model predictions; thus supporting decisions made by educators and psychologists. The framework was tested against a sample of 1248 student records including academic, behavioral and cognitive indicators. Experimental results demonstrated an accuracy of 91.45%, precision of 90.82%, recall of 90.17%, F1-score of 90.48%, and an AUC-ROC of 0.94. The performance improvements showed to be 6.1%, 4.3% and 2.8% compared to the Decision Tree, Random Forest and the LSTM models, respectively. It is believed that the proposed framework will lead to better outcomes in education through personalized intervention, with an accurate, interpretable and scalable solution for identifying learning challenges in the early stages of ADHD.