Fusion-Based Deep Learning Approach for Comprehensive Human Behavior Analysis from Visual Cues

Facial recognition is an essential component of computer vision, with applications in healthcare, behavioural science, security, and human–computer interaction. However, most existing techniques treat age estimation, emotion recognition, and personality prediction as separate problems, overlooking the natural relationships among these attributes. This study addresses that gap by introducing a unified Multi-Task Learning (MTL) and fusion-based framework capable of predicting all three characteristics simultaneously from facial data. The proposed model integrates three benchmark datasets: UTKFace for age estimation, FER2013 for emotion recognition, and the ChaLearn First Impressions dataset for personality analysis. Comprehensive preprocessing ensured data balance, consistency, and compatibility across tasks. The architecture employs Convolutional Neural Networks (CNNs) enhanced with attention mechanisms and transfer learning to extract high-quality feature representations. A combination of classification, regression, and ensemble learning methods was used to improve predictive accuracy and model robustness. Experimental results demonstrate that the MTL model outperforms individual single-task models. The system achieved 92.14% accuracy (F1 = 0.9213) in age-group classification, a real-age estimation performance of MAE = 5.40 years with R2 = 0.91, and 69% accuracy for seven-class emotion recognition. Personality prediction also showed strong reliability. A fusion model further boosted the overall predictive power to 88.97%. Beyond performance, the research emphasizes ethical considerations, including privacy protection, bias reduction, and informed consent. These measures promote responsible and fair deployment of facial analysis technologies. The findings highlight the potential of holistic, ethically aligned facial analysis systems for applications in medical diagnostics, personalized e-learning, adaptive interfaces, and behavioural assessment.