VOLUME 17 (Supplement)

SciEnggJ%202024%20Special%20Issue%201 7 Pasham%20et%20al

SciEnggJ 17 (Supplement) 096-102
available online: March 27, 2024
DOI: https://doi.org/10.54645/202417SupZAM-51

*Corresponding author
Email Address: meghabansal05@gmail.com
Date received: September 4, 2023
Date revised: December 30, 2023
Date accepted: January 15, 2024


An evolutionary and insightful graph neural network based hybrid model for deciphering tenacious stress detection of humans using facial emotion recognition: coinage to affective computing

Megha Bansal* and Vaibhav Vyas

Department of Computer Science, Banasthali Vidyapith, Rajasthan, India

KEYWORDS: stress, mental health, machine learning, affective computing

In the realm of affective computing, facial emotion recognition plays a pivotal role in ensuring prooper stres detection in humans. Within the domain of facial emotion recognition, accurate detection of emotions from facial expressions holds immense significance. Existing methods often neglect intricate connections between various facial features, potentially leading to challenges in discerning naunced emotional states. In response to this, we introdue approach termed Graph Neural Network-based Facial Emotion Recognition. Our technique harnesses the capabilities of graph neural network (GNNs) to surmount these limitations. The process inititaes with the creation of a graph that encapsulates the similarity between individual facial expression samples. Subsequently, this graph is fed into the model for intricate feature mapping. The outputs generated by this integrated model ingeniously merge the feature insights from adjacent facial expression samples, thereby enhancing the representation of data for downstream emotion recognition tasks. The graph neural network model processed facial expression samples are then directed to an emotion adjacency matrix that discerns various emotional states. The final outcomes are the culmination of results from this classification process. By accurately defining the top samples exhibiting the most pronounced emotional cues, our method efficiency determines the prevailing emotional state.Through comprehensive experimentation involving generation of 99.89% accuracy on publicily available facial expression dataset, our results highlight this algorithms superiority. This approach showcases a substantial increase in recognition accuracy, outperforming the closest competing algorithms by a significant margin. This demonstrates the efficacy and promise of this algorithm in revolutionizing stress detection, potentially ushering in a new era of precise and nuanced stress analysis form facial expressions.

© 2024 SciEnggJ
Philippine-American Academy of Science and Engineering