Bridging Accuracy and Efficiency in Facial Expression Recognition: An Adaptive Swarm Intelligence Framework for Resource-Constrained Environments

Facial expression recognition (FER) systems confront major computational hurdles due to high-dimensional feature spaces, limiting their application on resource-constrained devices. This research offers a unique Particle Swarm Optimization with Adaptive Opposition-Based Learning (PSO-AOBL) framework for optimal feature selection in FER systems. The suggested adaptive technique prevents premature convergence while preserving quick optimization performance by dynamically initiating opposition-based learning based on swarm diversity. Three benchmark datasets—CK+, FER2013, and JAFFE—are used to assess our methodology. Experimental results show PSO-AOBL reduces features by 73% (from 3,918 to 1,050 features) and inference time by 75% while achieving 94.2% accuracy on the CK+ dataset. Compared to normal PSO, our technique offers 2.1% accuracy improvement and 43% faster convergence.

• Facial expression recognition
• feature selection
• particle swarm optimization
• swarm intelligence
• computational efficiency

• Shweta Sharma, Research Scholar, College of Computing Science & Information Technology Teerthanker Mahaveer University, Moradabad-244001, Uttar Pradesh, India
• Prof. Ashok Kumar, Professor, College of Computing Science & Information Technology Teerthanker Mahaveer University, Moradabad-244001, Uttar Pradesh, India

Sharma and A. Kumar, “Bridging accuracy and efficiency in facial expression recognition: An adaptive swarm intelligence framework for resource-constrained environments,” IPEM Journal of Computer Application & Research, vol. 10, pp. 13–19, Dec. 2025, doi: