Attention-Deficit/Hyperactivity Disorder (ADHD) is a prevalent neurodevelopmental condition that usually starts in childhood and has overlapping symptoms with disorders such as depression and oppositional defiant disorder, hence challenging diagnosis. This work investigates the potential for Machine Learning (ML) to enhance diagnostic accuracy. We worked with data from the Healthy Brain Network (HBN), with 1,214 participants and merging diagnostic, socio-demographic, emotional, parenting, and functional MRI (fMRI) data. The brain connectome data alone includes around 20,000 features, requiring dimensionality reduction techniques like Principal Component Analysis (PCA), Autoencoders, Self-Organizing Maps (SOMs) with U-Matrix visualization, and feature engineering to manage complexity. We trained several machine learning models to predict ADHD and gender. Since the dataset was imbalanced, particularly for ADHD cases, we applied techniques like Synthetic Minority Oversampling Technique (SMOTE) to improve model fairness. When tested on 303 individuals, our models achieved 84% accuracy for ADHD prediction and about 70% for gender. These results show that combining diverse data sources with machine learning can significantly enhance the reliability of ADHD diagnosis.