<![CDATA[Emotion recognition is a fundamental aspect of affective computing, focusing on identifying and interpreting human emotional states. Among various modalities, facial emotion recognition has gained significant attention due to its non-intrusive nature and extensive applicability across domains such as e-learning, healthcare, marketing, e-commerce, and psychology. A wide range of approaches has been employed to address the challenges inherent in facial emotion classification. There remains a lack of a holistic, structured framework that critically evaluates both the advantages and shortcomings of deep networks while introducing attention-based and Transformer-driven models. Therefore, to address this gap, this paper presents a systematic review of peer-reviewed FER studies of deep learning models published between 2022 and 2025. This paper presents the study of advanced deep learning architectures for facial emotion detection, emphasizing the predominance of Deep Learning models including Transformer-based architectures, hybrid CNN–Transformer models, spatiotemporal learning approaches, and novel attention mechanisms. This research work provides analysis of deep learning model architectures, learning strategies, datasets, evaluation protocols, and performance metrics reported in state-of-the-art FER research. It identifies common issues including computational complexity, real-world robustness, generalization across datasets, and data imbalance. It also analyzes current research challenges, limitations and their practical significance. Furthermore, this research work identified and discussed the possible opportunities, unresolved issues of human facial emotion recognition and provided the future directions. The objective of this study is to provide actionable insights for researchers and practitioners, guiding future research toward more robust, accurate, and interpretable FER systems.]]>
