Functional MRI reveals brain network organization and, combined with socio-demographic and behavioral metadata, can aid in identifying ADHD, distinguishing sex, and estimating age. Using precomputed fMRI connectivity matrices and rich metadata (e.g., handedness, parental education, behavioral scores, parenting style), we developed an interpretable machine-learning workflow evaluated via five-fold cross-validation. Models reliably discriminated ADHD (F1 = 0.85 ± 0.02), achieved moderate sex classification (F1 = 0.62), and predicted age with low error (RMSE 0.616; R2 0.618). This approach highlights the benefit of integrating neuroimaging and metadata for diverse predictive tasks, emphasizing transparency and robust validation over model complexity. While not for clinical use, it illustrates the potential of data-driven methods to study neurodevelopmental differences and guide future biomarker research.