Facial age estimation using a hybrid CNN–LSTM architecture on the UTKFace dataset

Facial age estimation plays an important role in biometrics, healthcare analytics, and human–computer interaction systems. While convolutional neural networks (CNNs) have demonstrated strong capability in extracting spatial facial features, modeling higher-level feature dependencies remains a challenge. This study proposes a hybrid CNN–LSTM architecture for facial age estimation using the UTKFace dataset. The CNN component extracts hierarchical spatial representations from aligned facial images, and the LSTM layer models inter-feature dependencies to enhance regression performance. Age prediction is formulated as a regression task and evaluated using Mean Absolute Error (MAE), Root Mean Square Error (RMSE), and R² metrics. The CNN-LSTM hybrid models were then found to be more effective for age estimation in experimental results than the baseline deep learning models. The model successfully retrieved hierarchical representations of faces, and modeled relationships between facial features, leading to better age prediction and smaller prediction error. The findings highlight the effectiveness of integrating sequential modeling mechanisms into spatial feature learning frameworks for improved age estimation.