Predicting traits and behaviors from brain imaging data is an increasingly important area of neuroscience that can offer valuable insights into brain organization and functional mechanisms. Connectome-based predictive modeling (CPM) is a widely used approach that outperforms several traditional methods; however, its reliance on linear predictive algorithms may limit its ability to capture complex relationships. This study introduces novel models incorporating advanced regression algorithms to address this limitation. Using data from the Human Connectome Project, decision tree-based models, specifically Random Forest and AdaBoost, are evaluated against traditional linear regression. Results demonstrate that these non-linear models achieve superior predictive performance on resting-state and social settings data. These findings highlight the potential of alternative algorithms to enhance CPM and contribute to improved predictive accuracy in computational neuroscience.