Machine learning methods applied to resting-state functional MRI (rs-fMRI) data can reveal brain network alterations associated with psychiatric disorders. This study presents an interpretable classification pipeline combining Support Vector Classification, Random Forest, and SHapley Additive exPlanations for feature interpretability. Functional connectivity matrices were derived from the UCLA Consortium for Neuropsychiatric Phenomics dataset using Yeo’s 7- and 17-network cortical parcellations and Pearson, Spearman, and Kendall correlation metrics. After preprocessing with fMRIPrep and confound regression, models were optimized using Optuna-based hyperparameter search and evaluated on held-out test sets. The SVC model outperformed RF across most configurations, achieving a maximum accuracy of 0.622 using the 17-network parcellation and Pearson correlation. Retraining SVC on SHAP-selected features improved accuracy to 0.735 and enhanced interpretability by identifying limbic, salience, control, and default mode network interactions as the most discriminative connections. These findings demonstrate that integrating explainable AI with functional connectivity modeling provides biologically meaningful insights into psychiatric neurodysfunction while maintaining robust predictive performance.