Emotion-Driven Human-Computer Interaction: A Multi-Modal Facial Recognition System for Age, Gender, and Emotion Detection using Hybrid Algorithm
Computer and Decision Making: An International Journal · 04.11.2026 · 3 (6S) · 3616-3631
DOI: 10.1922/comdem.v3i6s.19108 ↗Kurz gefasst
Fachartikel in der Zeitschrift Computer and Decision Making: An International Journal, online erschienen am 04.11.2026. Verfasst von Sree Lakshmi P und 4 weiteren Autor:innen. Der Volltext ist frei zugänglich (Open Access, CC BY 4.0).
Abstract (Originalsprache)
Emotion-aware human–computer interaction (HCI) requires systems that can read affective and demographic cues from the face quickly, on commodity hardware, and without sending images to the cloud. This paper presents a unified multi-task framework for emotion recognition, gender classification and age estimation from facial images, built around a compact convolutional neural network (CNN) whose feature map is scanned row-wise by a bidirectional long short-term memory (BiLSTM) layer. We evaluate the framework against a histogram-of-oriented-gradients support vector machine (HOG+SVM) and a recurrent (GRU) baseline on two public benchmarks, FER-2013 (35,887 images, seven emotions) and UTKFace (23,705 images), using a strictly controlled protocol: identical splits, a single CPU core, a six-epoch training budget, 2,000-sample bootstrap confidence intervals and exact McNemar tests. The hybrid network significantly outperforms the plain CNN on emotion recognition (48.2% vs. 47.0%, p = 0.016) and on gender classification (83.4% vs. 78.1%, p < 0.001), and it cuts the age error for the oldest group from 27.6 to 14.6 years. However, under this short budget the classical HOG+SVM remains the most accurate model (58.3% emotion accuracy, 87.8% gender accuracy, 8.00-year age MAE), a finding that we report openly rather than obscure. Robustness tests show that all neural models are fragile to additive noise, and a subgroup analysis reveals accuracy gaps linked to predicted gender and age that are largely explained by class composition. We talk about ethical boundaries, including privacy, consent, demographic bias and the fact that facial expression is not a reliable indicator of emotion, and share code to recreate all the numbers.
Zitieren
APA
P, S. L., Roy, R., O, P. M., Raj, A., & Santhosha, S. (2026). Emotion-Driven Human-Computer Interaction: A Multi-Modal Facial Recognition System for Age, Gender, and Emotion Detection using Hybrid Algorithm. Computer and Decision Making: An International Journal, 3(6S), 3616-3631. https://doi.org/10.1922/comdem.v3i6s.19108
Vancouver
P SL, Roy R, O PM, Raj A, Santhosha S. Emotion-Driven Human-Computer Interaction: A Multi-Modal Facial Recognition System for Age, Gender, and Emotion Detection using Hybrid Algorithm. Computer and Decision Making: An International Journal. 2026;3(6S):3616-3631. doi:10.1922/comdem.v3i6s.19108
BibTeX
@article{P2026,
author = {P, Sree Lakshmi and Roy, Ratnakirti and O, Pallavi M and Raj, Anushree and Santhosha, Soumya},
title = {Emotion-Driven Human-Computer Interaction: A Multi-Modal Facial Recognition System for Age, Gender, and Emotion Detection using Hybrid Algorithm},
journal = {Computer and Decision Making: An International Journal},
year = {2026},
volume = {3},
number = {6S},
pages = {3616--3631},
doi = {10.1922/comdem.v3i6s.19108}
}
EUVIMED