In the information age, vast data streams from fields such as cybersecurity, social media analysis, finance, and energy present challenges, particularly concept drift, where data properties change over time, threatening predictive models’ effectiveness. This study introduces PAWS, a novel passive concept drift adaptation approach based on instance weighting and subspace alignment. PAWS addresses concept drift by adjusting the source domain's weight and learning the subspace transformation between source and target domains. Theoretical analysis of PAWS's generalization boundary highlights its robustness. Comparative analysis and experimental evaluation demonstrate that PAWS outperforms existing methods in detecting concept drift in both real and synthetic data streams, offering effective solutions for maintaining model performance in dynamic environments.