Transactions on Machine Intelligence 2821-1693 8 2 2025 06 01 Leveraging Bagging Ensemble Architectures for the Automated Diagnosis of Dyslexia via Visual Task Paradigm Analysis 101 110 244851 10.22034/tmi.2025.244851 EN M. Zaree Department of Biomedical Engineering, K. N. Toosi University of Technology, Tehran, Iran B. Arefi-Rad Department of Electrical Engineering, K. N. Toosi University of Technology, Tehran, Iran M. Mohebbi Department of Biomedical Engineering, K. N. Toosi University of Technology, Tehran, Iran R. Rostami Department of Psychology, University of Tehran, Tehran, Iran Journal Article 2025 02 13 Dyslexia is a neurobiological learning disability that fundamentally impairs a child’s literacy acquisition, specifically manifesting as persistent deficits in reading and writing. Absent a timely diagnosis, the disorder precipitates profound psychological distress and academic marginalization for both the pediatric patients and their families. Furthermore, delayed intervention often results in cumulative achievement gaps that become increasingly difficult to bridge by secondary education. Consequently, early screening and clinical intervention are paramount to preserving student self-esteem and optimizing long-term academic trajectories. This study proposes an automated diagnostic framework utilizing a Bagging (Bootstrap Aggregating) ensemble learning approach to classify dyslexia in children. The methodology involves the rigorous preprocessing of electroencephalogram (EEG) signals recorded across a 19-channel montage. Feature extraction focused on the morphometry of Event-Related Potentials (ERPs), specifically quantifying the amplitude and latency of key components. To address the "curse of dimensionality" inherent in the high-dimensional feature set, Principal Component Analysis (PCA) was implemented for optimal feature reduction. To ensure the generalizability of the model and mitigate the risk of overfitting, a K-fold cross-validation strategy was employed during the training phase. Finally, the Bagging classifier was deployed to distinguish between dyslexic and neurotypical subjects. The proposed ensemble framework demonstrated robust performance, yielding an average classification accuracy of 90.6%. Notably, the model achieved a sensitivity rate of 100%, ensuring no dyslexic cases were omitted, and a specificity of 81.2%, reflecting its capability to accurately identify neurotypical controls.