AI in elderly care: Real-time emotion analysis for better well-being

Facial expression recognition (FER) has been widely used in healthcare, assistive technologies, and emotion-aware AI systems, yet age-related changes in facial muscles and skin elasticity make it significantly harder for AI models to recognize emotions in older adults.

As the global population ages, healthcare and emotional well-being of the elderly have become critical areas for innovation. With estimates predicting that the number of people aged 65 and older will surpass 2.2 billion by the late 2070s, the demand for smart, non-invasive monitoring technologies is higher than ever. One promising area is facial expression recognition (FER), which uses artificial intelligence (AI) to assess emotions, mental health conditions, and well-being among seniors. However, despite the rapid progress of deep learning in computer vision, FER systems have traditionally been biased toward younger faces, limiting their accuracy and real-world applicability for elderly populations.

A recent study, "Deep Learning-Based Facial Expression Recognition for the Elderly: A Systematic Review", authored by F. Xavier Gaya-Morey, Jose M. Buades-Rubio, Philippe Palanque, Raquel Lacuesta, and Cristina Manresa-Yee, explores this challenge. Available in arXiv (2025), the research reviews 31 FER studies conducted over the past decade, analyzing dataset limitations, algorithmic challenges, and real-world deployment hurdles for AI-based facial emotion recognition in the elderly. The study emphasizes the need for age-diverse datasets, privacy-preserving AI solutions, and explainable AI (XAI) methods to bridge the gap between academic research and practical applications in elderly care.

Challenges in facial expression recognition for the elderly

Facial expression recognition (FER) has been widely used in healthcare, assistive technologies, and emotion-aware AI systems, yet age-related changes in facial muscles and skin elasticity make it significantly harder for AI models to recognize emotions in older adults. The study highlights three major challenges that hinder accurate FER for seniors:

First, a lack of elderly-specific datasets limits model performance. Most commonly used FER datasets - such as FER-2013 and AffectNet - underrepresent elderly individuals, leading to age bias in AI predictions. Additionally, existing datasets often have class imbalances, where expressions such as happiness are overrepresented while subtler emotions like sadness or confusion are underrepresented, making AI models less effective in detecting negative emotional states in seniors.

Second, deep learning models struggle with age-induced facial changes. While convolutional neural networks (CNNs) dominate FER research, they often fail to generalize well to wrinkled faces, lower facial muscle activity, and subtle expressions found in elderly individuals. This leads to misclassification of emotions and reduced model accuracy when applied to older populations.

Third, real-world deployment remains limited, mainly due to privacy concerns, computational constraints, and lack of explainability. Many AI-based healthcare solutions require constant camera monitoring, raising concerns about data security and ethical AI use. Moreover, deep learning models are often black-box systems, making it difficult for caregivers or doctors to trust AI-generated emotional assessments without a clear explanation of the model’s decisions.

Advancements in AI for elderly FER

To address these challenges, researchers have begun integrating new AI techniques and diverse datasets into FER systems for the elderly. The study identifies three key advancements driving improvements in AI-based emotion recognition:

First, multimodal AI approaches are enhancing FER accuracy by combining facial expressions with voice, text, and physiological data. Studies have shown that adding speech emotion analysis or biometric signals can improve AI’s ability to detect emotional distress in elderly individuals, especially those with cognitive impairments or speech limitations.

Second, lightweight deep learning models are making FER more accessible for resource-constrained environments, such as nursing homes, mobile health applications, and assistive robots. Models like MobileNet and mini-Xception have been successfully deployed on low-power devices, making real-time emotion recognition feasible in elderly care settings.

Third, explainable AI (XAI) techniques are improving transparency and trust in FER systems. By incorporating LIME (Local Interpretable Model-Agnostic Explanations) and Grad-CAM (Gradient-weighted Class Activation Mapping), researchers can visualize which facial features AI models rely on for emotion classification, allowing caregivers to validate AI predictions before making healthcare decisions.

Ethical and practical considerations for real-world deployment

Despite AI’s potential to revolutionize elderly emotion monitoring, real-world implementation requires addressing privacy, usability, and bias mitigation. The study highlights several key factors that must be considered before FER technology can be widely adopted in elderly care:

Privacy remains a critical challenge. Many elderly individuals are uncomfortable with constant video monitoring, making it essential to develop privacy-preserving AI techniques. Some studies suggest using facial landmark detection instead of raw facial images to anonymize user data while still allowing AI to detect emotions. Others propose on-device AI processing, where emotion recognition happens locally on mobile devices or robots, eliminating the need for cloud-based data storage.

Economic cost is another major barrier to adoption. While AI-powered FER has demonstrated effectiveness in research settings, real-world deployment requires affordable, scalable solutions. Researchers suggest using low-cost cameras, energy-efficient deep learning models, and edge AI processing to make FER technology more accessible in nursing homes, assisted living facilities, and home healthcare systems.

Ethical AI and bias reduction are essential for fairness. Since many AI models underperform on elderly faces, researchers emphasize the need for age-balanced training datasets, algorithmic fairness audits, and user-centered AI design. By ensuring equitable AI performance across age groups, FER systems can provide more reliable emotion monitoring without reinforcing age-related biases.

The future of AI-driven elderly care

The study underscores the transformative potential of deep learning-based facial expression recognition in elderly care, but also highlights the barriers that must be overcome for real-world impact. While multimodal AI, lightweight models, and explainable AI techniques have made significant progress in improving FER accuracy, challenges related to privacy, affordability, and bias mitigation still hinder widespread adoption.

Moving forward, the researchers advocate for interdisciplinary collaboration between AI developers, healthcare professionals, and policymakers to ensure that FER systems are ethical, transparent, and accessible. The integration of AI into elderly care must prioritize user trust, data security, and practical usability, ensuring that facial expression recognition becomes a reliable tool for improving emotional well-being and quality of life for aging populations.

By addressing these challenges, AI-powered FER systems could revolutionize elderly healthcare, providing caregivers with real-time emotional insights, enabling early detection of mental health conditions, and enhancing human-AI interaction in assistive technologies. The study serves as a call to action for researchers and policymakers to bridge the gap between AI innovation and real-world elderly care solutions, ensuring that no one is left behind in the age of artificial intelligence.