<p>Domain adaptation has become a critical area of research to improve model performance across varying domains by transferring information obtained from a source domain to a target domain. Despite that, existing methods often struggle due to feature misalignment, domain shift, and label scarcity. In this research work, a novel Domain-generalized semi-supervised learning model is introduced to resolve those limitations through a synergistic integration of Deep Neural Networks (DNN) with attention mechanisms, adversarial training, and semi-supervised learning. The DNN with attention mechanisms is introduced to derive domain-invariant features and minimize domain-specific variations which contribute to better feature alignment and enhanced stability in adversarial training. An adversarial domain adaptation mechanism is introduced to minimize domain shift and allows the architecture to generate robust feature representations that exhibit good generalizability across domains. The semi-supervised learning improves the robustness of the method in circumstances with limited labeled data by avoiding unnecessary feature alignment and maintaining feature consistency. The comprehensive experiments on the DomainNet, Office-31, and Digits datasets show significant improvements in accuracy, Domain Adaptation Accuracy (DAA), and Maximum Mean Discrepancy (MMD) and surpassed baseline methods by achieving 98.37% accuracy, 0.95 DAA, and 0.023 MMD, demonstrating better generalizability and adaptability in domain adaptation. The ablation experiment and qualitative analysis further confirm the superiority of the approach to establish a solid foundation for real-world applications.</p>