The proliferation of deep learning models across critical domains increases the need to have explainable artificial intelligence (XAI) systems that are transparent and understandable in their decision-making process. Attention mechanisms, initially meant to improve the performance of models in sequence-to-sequence tasks, have been shown to be promising intrinsic explainability methods that provide information about the way models reason without the need to analyse them post-hoc. This systematic review investigates the applications, effectiveness, and limitations of attention-based explainability in computer vision, natural language processing, medical diagnostics, and time-series analysis. We examined 68 peer-reviewed research papers published in 2017 to 2025 assessing attention mechanisms on explainability measures such as faithfulness, plausibility, and robustness. Spatial attention mechanisms demonstrate better explainability scores (faithfulness: 0.84, plausibility: 0.82, robustness: 0.75), and healthcare uses show strong performance (96.1% accuracy, 0.85 faithfulness). Comparative analysis shows that attention-based methods possess computational benefits over LIME, SHAP, and Grad-CAM. Challenges include changeability of attention under perturbations (27.9%), prediction variance, and non-homogeneous evaluation patterns; robustness (42.6%) and human evaluation (35.3%) proportions were low. We propose future research should focus on causal attention, explainable models, adaptive system designs, and standardized evaluation frameworks.

Yuanyumeng Zhu. Applications of Attention Mechanisms in Explainable Machine Learning. Academic Journal of Computing & Information Science (2025), Vol. 8, Issue 10: 50-62. https://doi.org/10.25236/AJCIS.2025.081008.

[1] M. Mersha, K. Lam, J. Wood, A. K. AlShami, and J. Kalita, “Explainable artificial intelligence: A survey of needs, techniques, applications, and future direction,” Neurocomputing, vol. 599, Sep. 2024, doi: 10.1016/J.NEUCOM.2024.128111.

[2] B. Kotipalli, “The Role of Attention Mechanisms in Enhancing Transparency and Interpretability of Neural Network Models in Explainable AI,” Harrisburg University Dissertations and Theses, Apr. 2024, Accessed: Nov. 12, 2025. [Online]. Available: https://digitalcommons.harrisburgu.edu/dandt/2

[3] A. Vaswani et al., “Attention Is All You Need,” p. 1, Jun. 2017, Accessed: Nov. 12, 2025. [Online]. Available: https://arxiv.org/pdf/1706.03762

[4] L. Hu, Y. Liu, N. Liu, M. Huai, L. Sun, and D. Wang, “SEAT: Stable and Explainable Attention,” Proceedings of the AAAI Conference on Artificial Intelligence, vol. 37, no. 11, pp. 12907–12915, Jun. 2023, doi: 10.1609/AAAI.V37I11.26517.

[5] G. Brauwers and F. Frasincar, “A General Survey on Attention Mechanisms in Deep Learning,” IEEE Trans Knowl Data Eng, vol. 35, no. 4, pp. 3279–3298, Apr. 2023, doi: 10.1109/TKDE. 2021. 3126456.

[6] A. Mueed Hafiz et al., “Attention mechanisms and deep learning for machine vision: A survey of the state of the art,” Jun. 2021, Accessed: Nov. 12, 2025. [Online]. Available: https://arxiv.org/pdf/2106. 07550

[7] M. Hassanin, S. Anwar, I. Radwan, F. S. Khan, and A. Mian, “Visual attention methods in deep learning: An in-depth survey,” Information Fusion, vol. 108, p. 102417, Aug. 2024, doi: 10.1016/J. INFFUS.2024.102417.

[8] A. M. Salih et al., “A Perspective on Explainable Artificial Intelligence Methods: SHAP and LIME,” Advanced Intelligent Systems, vol. 7, no. 1, p. 2400304, Jan. 2025, doi: 10.1002/AISY. 202400304; WGROUP:STRING:PUBLICATION.

[9] V. Vimbi, N. Shaffi, and M. Mahmud, “Interpreting artificial intelligence models: a systematic review on the application of LIME and SHAP in Alzheimer’s disease detection,” Brain Informatics 2024 11:1, vol. 11, no. 1, pp. 1–29, Apr. 2024, doi: 10.1186/S40708-024-00222-1.

[10] S. Nerella et al., “Transformers and large language models in healthcare: A review,” Artif Intell Med, vol. 154, p. 102900, Aug. 2024, doi: 10.1016/j.artmed.2024.102900.

[11] A. Parvaiz, M. A. Khalid, R. Zafar, H. Ameer, M. Ali, and M. M. Fraz, “Vision Transformers in medical computer vision—A contemplative retrospection,” Eng Appl Artif Intell, vol. 122, p. 106126, Jun. 2023, doi: 10.1016/J.ENGAPPAI.2023.106126.

[12] V. Padmavathi and K. Ganesan, “Metaheuristic optimizers integrated with vision transformer model for severity detection and classification via multimodal COVID-19 images,” Scientific Reports 2025 15:1, vol. 15, no. 1, pp. 1–19, Apr. 2025, doi: 10.1038/s41598-025-98593-w.

[13] M. Zhang, Y. Zhang, S. Liu, Y. Han, H. Cao, and B. Qiao, “Dual-attention transformer-based hybrid network for multi-modal medical image segmentation,” Scientific Reports 2024 14:1, vol. 14, no. 1, pp. 1–22, Oct. 2024, doi: 10.1038/s41598-024-76234-y.