The integration of augmented reality (AR) and machine learning (ML) presents transformative opportunities for enhancing consumer experiences and decision-making processes. This study investigates how ML-driven AR features—specifically augmentation, interactivity, and vividness—affect users’ emotional responses and purchase intentions in virtual makeup trials. Guided by the Stimulus-Organism-Response (SOR) framework and the Pleasure-Arousal-Dominance (PAD) emotional model, we apply supervised learning algorithms to analyze and predict consumer emotions and behaviors based on real-time interaction data. Using structural equation modeling (SEM), we validate the relationships between emotional states and purchase intentions, revealing that AR features optimized through ML significantly boost emotional engagement. Notably, feelings of dominance and arousal serve as key mediators in enhancing user pleasure and influencing purchase decisions. These findings offer valuable insights for developers and marketers, demonstrating the potential of intelligent AR systems to create personalized, immersive consumer experiences in digital retail environments.