Comparative Analysis of a Simple CNN and Fine-Tuned ResNet50 for Deepfake Image Detection: Performance and Computational Efficiency Evaluation

Rifda Triani Mutmainah; Asti Herliana

doi:10.61255/decoding.v4i2.1629

Authors

Rifda Triani Mutmainah Universitas Adhirajasa Reswara Sanjaya
Asti Herliana Universitas Adhirajasa Reswara Sanjaya

DOI:

https://doi.org/10.61255/decoding.v4i2.1629

Keywords:

CNN, Computational efficiency, Deepfake, Fine-tuning

Abstract

Purpose – Deepfake-related cybercrime is an increasingly troubling cybersecurity threat, since such content can now be produced from a single facial photo using freely available face-swapping techniques, while human ability to distinguish real from fake images remains limited (48.2–59%). This study compares a simple CNN and a fine-tuned ResNet50 within an identical, controlled framework.
Methods – A controlled experiment compared a simple CNN (trained from scratch) and ResNet50 (two-phase fine-tuning) on 20,000 images from the Kaggle Deepfake and Real Images dataset (80:10:10 split, seed = 42). Candidate images were deduplicated before the data split; zero cross-split duplicates were confirmed. ResNet50 used ResNet-specific preprocessing; the CNN used inputs normalized to [0,1]. Evaluation used accuracy, precision, recall, F1-score, AUC-ROC, and specificity, with a paired McNemar's test as the primary significance measure.
Findings – The CNN outperformed ResNet50 on six of seven metrics (accuracy 90.90% vs. 89.80%; AUC-ROC 97.08% vs. 96.52%), while ResNet50 achieved higher recall (93.10% vs. 91.80%). The accuracy difference was not statistically significant (p = 0.200). The CNN was 69.5 times smaller and trained faster (12.2 vs. 16.5 minutes). Neither model showed clear overfitting.
Research implications – The findings rest on a single subset, a single split, and a single training run per model, limited to 20 epochs; the potential of ResNet50 with longer training has not been explored.
Originality – This study combines accuracy, generalization-related diagnostics, and efficiency in a single controlled comparison, applies a paired test appropriate for a shared test set, and filters duplicates before the data split.

Abstract views: 34 ,

PDF downloads: 13

Downloads

Download data is not yet available.

References

C. Zhai, S. Wibowo, and L. D. Li, “The effects of over-reliance on AI dialogue systems on students’ cognitive abilities: a systematic review,” Smart Learning Environments, vol. 11, no. 1, Jun. 2024, doi: 10.1186/S40561-024-00316-7.

J. M. Camacho, A. Couce-Vieira, D. Arroyo, and D. Rios Insua, “A Cybersecurity Risk Analysis Framework for Systems with Artificial Intelligence Components,” 2024, Accessed: Jul. 16, 2026. [Online]. Available: https://doi.org/10.48550/arXiv.2401.01630

V. E. Pattiradjawane et al., “Deepfakes as a New Cybersecurity Threat: A Literature Review of Risks to Digital Identity, Social Engineering, and Information Trust” ALGORITHM: Journal of Computer Science and Computational Intelligence, vol. 2, no. 1, pp. 19–30, May 2026, doi: 10.30598/ALGORITHM.V2I1.19-30.

M. S. Rana, M. N. Nobi, B. Murali, and A. H. Sung, “Deepfake Detection: A Systematic Literature Review,” IEEE Access, vol. 10, pp. 25494–25513, 2022, doi: 10.1109/ACCESS.2022.3154404.

V. Pirogov, “Visual Language Models as Zero-Shot Deepfake Detectors,” Jul. 2025, Accessed: Jul. 16, 2026. [Online]. Available: https://arxiv.org/pdf/2507.22469

S. D. Bray, S. D. Johnson, and B. Kleinberg, “Testing human ability to detect ‘deepfake’ images of human faces,” J. Cybersecur., vol. 9, no. 1, Jan. 2023, doi: 10.1093/CYBSEC/TYAD011.

S. N. Amartama, A. N. Hidayah, P. K. Sari, and R. A. Ramadhani, “Implementation of Convolutional Neural Networks (CNN) in Handwritten Pattern Recognition” Seminar Nasional Teknologi & Sains, vol. 3, no. 1, pp. 133–138, Jan. 2024, doi: 10.29407/STAINS.V3I1.4155.

A. Puspitasari, D. Sava Salsabila, and D. Roliawati, “Application of ResNet-50 CNN for Classification Optimization on Fashion Data” INDONESIAN JOURNAL ON DATA SCIENCE, vol. 3, no. 1, pp. 1–12, Jun. 2025, doi: 10.30989/IJDS.V3I1.1533.

L. Alzubaidi et al., “Review of deep learning: concepts, CNN architectures, challenges, applications, future directions,” J. Big Data, vol. 8, no. 1, Mar. 2021, doi: 10.1186/S40537-021-00444-8.

Z. N. Ashani et al., “Comparative Analysis of Deepfake Image Detection Method Using VGG16, VGG19 and ResNet50,” Journal of Advanced Research in Applied Sciences and Engineering Technology Journal homepage: Journal of Advanced Research in Applied Sciences and Engineering Technology, vol. 47, no. 1, pp. 16–28, 2025, doi: 10.37934/araset.47.1.1628.

Kusnaeni, I. R. Adriani, M. S. Hafid, A. Budi Andy B, and M. E. Rizal, “Deepfake Image Classification Using ResNet50 Feature Extraction and XGBoost Learning Model,” Journal of Mathematics, Computations and Statistics, vol. 8, no. 2, pp. 258–268, Sep. 2025, doi: 10.35580/JMATHCOS.V8I2.8387.

L. Guarnera et al., “The Face Deepfake Detection Challenge,” J. Imaging, vol. 8, no. 10, Oct. 2022, doi: 10.3390/JIMAGING8100263.

H. A. Pangestu and Kusrini, “Improving ResNet50 Architecture Performance to Address Overfitting in Skin Disease Classification” TEMATIK, vol. 11, no. 1, pp. 65–71, Jun. 2024, doi: 10.38204/TEMATIK.V11I1.1876.

Y. A. Chouhan, C. Chudasama, and D. K. Verma, “Deepfake Detection Using ResNet50: Performance Analysis on Celeb-DF Dataset,” Communications in Computer and Information Science, vol. 2426 CCIS, pp. 397–404, 2025, doi: 10.1007/978-3-031-86296-0_28.

E. H. I. Eliwa, “Enhancing Skin Cancer Diagnosis Through Fine-Tuning of Pretrained Models: A Two-Phase Transfer Learning Approach,” Int. J. Breast Cancer, vol. 2025, no. 1, 2025, doi: 10.1155/IJBC/4362941.

L. Feng et al., “Adaptive multi-source domain collaborative fine-tuning for transfer learning,” PeerJ Comput. Sci., vol. 10, p. e2107, Jun. 2024, doi: 10.7717/PEERJ-CS.2107/SUPP-1.

D. R. Rochmawati and L. Maryani, “Deep Learning-Based ResNet-50 Transfer Learning Approaches for Pneumonia Detection from Chest X-Ray Images: With and Without Fine-Tuning,” Indonesian Journal of Health Research and Development, vol. 3, no. 3, pp. 146–153, Nov. 2025, doi: 10.58723/IJHRD.V3I3.507.

J. Mallet, L. Pryor, R. Dave, and M. Vanamala, “Deepfake Detection Analyzing Hybrid Dataset Utilizing CNN and SVM,” ACM International Conference Proceeding Series, pp. 7–11, Jan. 2023, doi: 10.1145/3596947.3596954.

B. Cavia, E. Horwitz, T. Reiss, and Y. Hoshen, “Real-Time Deepfake Detection in the Real-World,” Jun. 2024, Accessed: Jul. 16, 2026. [Online]. Available: https://arxiv.org/pdf/2406.09398

S. T. Erukude, V. C. Marella, and S. R. Veluru, “Fourier-Based GAN Fingerprint Detection using ResNet50,” Proceedings of the 9th International Conference on Inventive Systems and Control, ICISC 2025, pp. 971–976, Oct. 2025, doi: 10.1109/ICISC65841.2025.11187724.

M. Karki, “deepfake and real images.” Accessed: May 10, 2026. [Online]. Available: https://www.kaggle.com/datasets/manjilkarki/deepfake-and-real-images?resource=download

A. Pakala, M. M. Rahman, S. Memon, and T. Ahmed, “Comparative Evaluation of Deep Learning Models for Fake Image Detection,” May 2026, Accessed: Jul. 15, 2026. [Online]. Available: https://arxiv.org/pdf/2605.20971

Y. Dogan, “A New Global Pooling Method for Deep Neural Networks: Global Average of Top-K Max-Pooling,” Traitement Du Signal, vol. 40, no. 2, pp. 577–587, Apr. 2023, doi: 10.18280/TS.400216.

A. Angeline and H. Kusniyati, “Performance Comparison of VGG19, ResNet50, DenseNet121, and MobileNetV2 in Detecting Deepfake Images” CESS (Journal of Computer Engineering, System and Science), vol. 9, no. 2, pp. 397–411, Jul. 2024, doi: 10.24114/CESS.V9I2.58671.

D. Vaishnavi, S. Jegan, J. Ganesh, and S. Sundaram, “Hybrid Deep Learning Approach for Deepfake Detection Using ResNet50 and EfficientNetB0,” Journal of Innovative Image Processing, vol. 7, no. 3, pp. 622–638, Aug. 2025, doi: 10.36548/JIIP.2025.3.003.