Deep Learning-Based Facial Expression Recognition for the Elderly: A Systematic Review

Existing facial recognition systems aren't optimized for elderly people: The review found that current datasets used to train these systems lack adequate representation of older adults, meaning the technology may not accurately recognize emotions in the population it's intended to help. This is a critical gap since age-related changes in facial features and expressions differ from younger populations.

Privacy and transparency are major barriers to real-world use: Beyond technical performance, the research identified privacy concerns and the need for "explainable AI" (systems that can justify their decisions) as key obstacles preventing these emotion-monitoring technologies from being deployed in actual elderly care settings, even when they work well in laboratory conditions.

Current technology relies heavily on lightweight neural networks: The review found that convolutional neural networks—particularly smaller, resource-efficient versions—dominate the field, which is practical for deployment in assisted living facilities and homes with limited computing power, but this focus may come at the cost of accuracy.

Combining multiple types of data could improve reliability: The researchers recommend integrating multimodal solutions (such as combining facial expressions with voice tone or body language) rather than relying on facial recognition alone, which could make these systems more robust and trustworthy for monitoring elderly emotional well-being.

The gap between research and real-world implementation remains significant: While academic progress has been made, the review emphasizes that very few systems have moved from research labs into actual elderly care environments, suggesting that technical advancement alone isn't sufficient without addressing ethical, privacy, and practical deployment challenges.