Nic Leander A. Delgado, Ziggy Delf R. Sy, Christian V. Maderazo
Attention-Deficit/Hyperactivity Disorder (ADHD) remains one of the most commonly diagnosed neurodevelopmental conditions in adults, yet its diagnosis largely relies on subjective assessments prone to bias and inconsistency. With the growing demand for more objective and scalable approaches, recent research has explored the integration of physiological and behavioral data to improve diagnostic reliability. This study aims to develop an ADHD classification model by combining multimodal data sources-specifically heart rate variability (HRV) and actigraphy-based motor activity, as well as cognitive performance through the Continuous Performance Test (CPT). Using the HYPERAKTIV dataset, the research preprocesses and engineers features from each modality before applying ensemble learning algorithms, which include Random Forest, Support Vector Machine, and Logistic Regression. These models are trained and validated using stratified 10-fold cross-validation. Feature normalization, selection, and synthetic minority oversampling (SMOTE) were applied to manage the data imbalance and overfitting risks. Experimental results demonstrated that the proposed Tri-Modal Fusion achieved a mean accuracy of 87.10% and a recall of 89.17%, significantly outperforming individual modalities. These findings confirm that fusing cognitive markers with physiological signals establishes a robust and objective screening tool for adult ADHD. © 2026 IEEE.
Department of Computer, Information Sciences, and Mathematics, University of San Carlos, Cebu City, Philippines
