A Hybrid CNN-BiLSTM-Attention Model for ADHD Detection Using Heart Rate Variability

Accurate detection of Attention-Deficit/Hyperactivity Disorder (ADHD) using physiological data presents a promising avenue for improving clinical diagnostics. This study proposes a hybrid deep learning model integrating Convolutional Neural Networks (CNN), Bidirectional Long Short-Term Memory (BiLSTM) networks, and an attention mechanism to classify ADHD using heart rate variability (HRV) data. The model combines both engineered and automatically extracted features to enhance predictive performance. A stratified 10-fold cross-validation approach was used for model training, and evaluation was conducted on a hold-out test set using key metrics including accuracy, precision, recall, F1-score, and Matthews correlation coefficient (MCC). The hybrid model achieved the highest performance with an accuracy of 0.76, F1-score of 0.75, and MCC of 0.52, outperforming baseline models that used either only automated features or only engineered features. ROC and Precision-Recall curves further confirmed the model's robustness, both yielding an average AUC and average precision of 0.82. Additionally, experiments with varying HRV segment lengths revealed that a segment length of 1024 samples provided optimal performance. These findings demonstrate that fusing diverse feature types within a hybrid deep learning framework can significantly enhance ADHD detection from HRV data.