[Paper Review] Enhancing Psychologists' Understanding through Explainable Deep Learning Framework for ADHD Diagnosis

Abdul Rehman, Ilona Heldal+1 more|arXiv (Cornell University)|Jan 28, 2026

Attention Deficit Hyperactivity Disorder0 citations

TL;DR

This paper proposes HyExDNN-RNN, an explainable deep learning framework for ADHD detection and multi-class diagnosis, with SHAP and PFI explanations, achieving 99% binary accuracy and 94.2% multi-class accuracy on ADHD200 data.

ABSTRACT

Attention Deficit Hyperactivity Disorder (ADHD) is a neurodevelopmental disorder that is challenging to diagnose and requires advanced approaches for reliable and transparent identification and classification. It is characterized by a pattern of inattention, hyperactivity and impulsivity that is more severe and more frequent than in individuals with a comparable level of development. In this paper, an explainable framework based on a fine-tuned hybrid Deep Neural Network (DNN) and Recurrent Neural Network (RNN) called HyExDNN-RNN model is proposed for ADHD detection, multi-class categorization, and decision interpretation. This framework not only detects ADHD, but also provides interpretable insights into the diagnostic process so that psychologists can better understand and trust the results of the diagnosis. We use the Pearson correlation coefficient for optimal feature selection and machine and deep learning models for experimental analysis and comparison. We use a standardized technique for feature reduction, model selection and interpretation to accurately determine the diagnosis rate and ensure the interpretability of the proposed framework. Our framework provided excellent results on binary classification, with HyExDNN-RNN achieving an F1 score of 99% and 94.2% on multi-class categorization. XAI approaches, in particular SHapley Additive exPlanations (SHAP) and Permutation Feature Importance (PFI), provided important insights into the importance of features and the decision logic of models. By combining AI with human expertise, we aim to bridge the gap between advanced computational techniques and practical psychological applications. These results demonstrate the potential of our framework to assist in ADHD diagnosis and interpretation.

Motivation & Objective

• Motivate improved, reliable ADHD diagnosis that is transparent and interpretable for clinical psychologists.
• Develop an explainable hybrid deep learning model (HyExDNN-RNN) that supports binary ADHD detection and multi-class categorization.
• Incorporate feature reduction and XAI methods (SHAP, PFI) to provide interpretable diagnostic insights.
• Evaluate model performance on the ADHD200 dataset and assess interpretability for clinical use.

Proposed method

• Fine-tune a hybrid Deep Neural Network and Recurrent Neural Network (HyExDNN-RNN) for ADHD detection and interpretation.
• Select features using Pearson correlation coefficient to reduce dimensionality.
• Train and compare multiple models (RF, DT, EXGB, DNN, LSTM, LSTM-GRU, LSTM-RNN) and the HyExDNN-RNN on reduced features.
• Apply SHAP to obtain local feature explanations and PFI for global feature importance to interpret decisions.
• Use standard preprocessing: missing data imputation, label encoding, normalization.
• Evaluate using binary and multi-class classification metrics (accuracy, precision, recall, F1-score, ROC-AUC).

Figure 1 : Proposed XAI-based Machine and Deep Learning Framework for ADHD Diagnosis and Explainability

Experimental results

Research questions

• RQ1 Can the HyExDNN-RNN framework achieve high accuracy in binary ADHD detection and multi-class classification on ADHD datasets?
• RQ2 How do SHAP and PFI explanations illuminate feature importance and model decision logic in ADHD diagnosis?
• RQ3 What is the impact of Pearson-based feature selection on model performance and interpretability?
• RQ4 How do various models compare to HyExDNN-RNN in terms of accuracy and robustness for ADHD detection?

Key findings

• HyExDNN-RNN achieves 99% accuracy (binary) and 0.99 ROC-AUC on binary ADHD classification.
• HyExDNN-RNN achieves 94.2% accuracy, precision, recall, and F1-score (multi-class).
• SHAP identifies inattentive, hyperactive/impulsive ADHD index as highly influential features for ADHD predictions.
• PFI provides global feature importance to reveal overall model behavior across predictions.
• EXGB and LSTM-RNN also perform well, but HyExDNN-RNN shows the strongest overall performance and interpretability.
• The framework demonstrates how AI with human expertise can support ADHD diagnosis and interpretation.

Figure 2 : Distribution of Dataset Classes. Typically Developing Children (TDC) Class 0, ADHD-combined (Class 1), ADHD-Hyperactive/Impulsive (Class 2), ADHD-Inattentive (Class 3)