REAL-TIME FACIAL EMOTION RECOGNITION USING MODIFIED EFFICIENTNETB0 WITH DUAL-DATASET TRAINING
Kashf Journal of Multidisciplinary Research · 27.03.2026 · 3 (03) · 122-148
DOI: 10.71146/kjmr879 ↗Kurz gefasst
Fachartikel in der Zeitschrift Kashf Journal of Multidisciplinary Research, online erschienen am 27.03.2026. Verfasst von Muhammad Nadeem und 5 weiteren Autor:innen. Der Volltext ist frei zugänglich (Open Access, CC BY 4.0).
Abstract (Originalsprache)
Recent advances in emotionally-aware computing have sparked growing interest in automated facial emotion recognition (FER), a field with far-reaching implications for human-computer interaction. While custom convolutional neural networks (CNNs) have demonstrated early promise in this domain, achieving consistently high performance under real-world conditions has remained a persistent challenge. This paper introduces an improved FER system built on a modified EfficientNetB0 architecture — pretrained on ImageNet and fine-tuned on a combined FER2013/CK+ dataset — designed to close the gap between laboratory accuracy and practical deployment. The proposed end-to-end pipeline integrates face detection, pre-processing, deep inference, and temporal smoothing into a seamless real-time workflow. Training with a 40-epoch cosine-annealing learning rate schedule yielded a peak validation accuracy of 88.4% on FER2013, representing a 14.1 percentage point improvement over the CNN baseline of 74.3%. Per-class recognition rates derived from the normalized confusion matrix are as follows: Disgust (96%), Surprise (95%), Angry (92%), Fear (91%), Happy (89%), Neutral (85%), and Sad (81%). Notably, the system achieves an inference speed of 22–28 frames per second on a standard laptop without GPU acceleration, underscoring its viability for real-world deployment in emotionally intelligent interaction applications.
Zitieren
APA
Nadeem, M., Rafi, H., Ihsan, M., Arslan, M., Chohan, S. R., & Akbar, W. (2026). REAL-TIME FACIAL EMOTION RECOGNITION USING MODIFIED EFFICIENTNETB0 WITH DUAL-DATASET TRAINING. Kashf Journal of Multidisciplinary Research, 3(03), 122-148. https://doi.org/10.71146/kjmr879
Vancouver
Nadeem M, Rafi H, Ihsan M, Arslan M, Chohan SR, Akbar W. REAL-TIME FACIAL EMOTION RECOGNITION USING MODIFIED EFFICIENTNETB0 WITH DUAL-DATASET TRAINING. Kashf Journal of Multidisciplinary Research. 2026;3(03):122-148. doi:10.71146/kjmr879
BibTeX
@article{Nadeem2026,
author = {Nadeem, Muhammad and Rafi, Hamza and Ihsan, Muhammad and Arslan, Muhammad and Chohan, Sohail Raza and Akbar, Wasif},
title = {REAL-TIME FACIAL EMOTION RECOGNITION USING MODIFIED EFFICIENTNETB0 WITH DUAL-DATASET TRAINING},
journal = {Kashf Journal of Multidisciplinary Research},
year = {2026},
volume = {3},
number = {03},
pages = {122--148},
doi = {10.71146/kjmr879}
}
EUVIMED